Good afternoon, and welcome to TuSimple's very first Investor Day. I'm Ryan Amerman, Senior Director of Investor Relations at TuSimple. On behalf of the entire TuSimple team, I wanna thank you for joining us in person in Tucson and virtually from around the world. Before we get started, I do have a disclaimer to read. I must let you know that we will be making forward-looking statements and presenting GAAP and non-GAAP measures during today's presentation. I ask you to review the disclaimer on the screen. Forward-looking statements can be identified by terms such as believe, intend, expect, or may, and you should not place undue reliance on forward-looking statements. Actual results can differ materially from forward-looking statements, and we do not undertake any obligation to update these forward-looking statements.
For more information about factors that may cause actual results to differ materially from forward-looking statements, please refer to our filings made with the SEC, which are also available on our investor relations website. As you can see, we have a packed agenda for you today. You'll have the opportunity to hear from Xiaodi, Jim, Pat, one of our very important partners, as well as several members of our technology team. With that, let's kick it off with a short welcome video. Let's try that again.
At TuSimple, we are transforming the $4 trillion global truck freight market.
Our proprietary revolutionary technology will make freight transportation safer, more eco-friendly, and more efficient.
We work with partners throughout the freight ecosystem, including hardware partners, service partners, carrier partners, and fleet customers.
We are a diverse team made up of engineers, scientists, strategists, leaders, drivers, and more, all united together in our shared vision.
Welcome to TuSimple's 2022 Investor Day. We are glad you're here with us, whether in person or virtually.
We are excited to tell you more.
About TuSimple and our leadership in autonomous trucking and the path ahead.
Welcome to the driver-out era.
Welcome to the driver-out era.
Welcome to the driver-out era.
Welcome to the driver-out era.
Welcome to the driver-out era.
Thank you, Ryan. Once again, thank you everyone for joining us in person in Tucson, and also virtually from around the world. I am Xiaodi Hou, the Chairman and CEO of TuSimple. On behalf of our entire team, we're very excited to share with you our story. At TuSimple, we are focused on addressing one of the very big problems since our founding. Our goal is to bring the most safe, fuel-efficient, and low-cost freight capacity to the market at scale. We're doing so by building what we believe is the most advanced Level 4 Autonomous Driving System for trucks, the most reliable hardware in conjunction with our OEM and Tier 1 partners, and combining those products with our unique and scalable go-to-market strategy, which we call the Autonomous Freight Network, or AFN. We believe our AFN would fundamentally reshape the estimated $4 trillion global freight transportation market.
Since our founding in 2025, we have been singularly focused on developing autonomous trucking. As a company, we have an illustrious history of achieving many industry firsts. In 2018, we were the first to demonstrate fully autonomous truck runs on both surface streets and highways. In 2020, we were the first to announce partnership with the major OEMs through our TRATON and Navistar partnerships. 2021 was a big year for us. Approximately one year ago, we become the first publicly listed autonomous driving company, raising $1 billion to fund our growth through commercialization. At the end of 2021, we became the first autonomous trucking company to demonstrate driver-out runs on open public roads and not relying on teleoperation. This is clear proof to our technology leadership.
To date, no other autonomous trucking company has been able to demonstrate the same. In TuSimple, leading the industry has always been in our DNA. Let's look at another short video to illustrate some of the industry firsts that our team has accomplished.
At TuSimple, we are proud to be the leader in autonomous trucking. Our company was founded in 2015 with a focus on revolutionizing the freight transportation industry through the deployment of artificial intelligence-based autonomous driving technology. Since our founding, we've accomplished many industry firsts, befitting our leadership status. First to demonstrate fully autonomous truck runs on surface streets and highways beginning in 2018, including a 70 mi zero disengagement run featured on 60 Minutes. First autonomous trucking company to receive investment from a major transportation carrier, UPS, in 2019. We've completed over 200,000 autonomous mi of paid freight haulage over the past three years for UPS and its North American freight forwarding services. Unveiled the TuSimple Autonomous Freight Network in 2020. The first transportation network of its kind, hauling paid freight for customers in the southwestern U.S.
First partnership program established in 2020 to deliver purpose-built Level 4 production semi-trucks together with Navistar. First autonomous vehicle company to go public via our traditional IPO in April 2021. First company in the world to operate a driver out semi-truck on open public roads beginning in December 2021. This is only the beginning. TuSimple, a better path forward.
Before we start diving into the details, let's lay the groundwork on who we are and the significant opportunity in front of us. We believe our total addressable market is $4 trillion globally. In the U.S. alone, which is our primary near-term focus, the truck freight market is $800 billion. It accounts for approximately 80% of the total freight transportation market in the U.S. There are over 2.3 million Class 8 semi-trucks on the road, and each year, these trucks travel over 175 billion mi delivering freights. The truck freight market is the backbone of what moves goods in the US as well as globally. We believe we have the right technology and business model to capture a meaningful share of this very large market.
The truck freight market is not only massive, but also highly concentrated along a few key freight routes. Within the middle mile, which is the target use of our technology, 80% of all goods are moved on just 10% of the nation's interstate highways. Some of our selected AFN core routes, which are the green highlighted routes on the map, represents approximately $10 billion-$13 billion of potential freight revenue alone. These are the routes that we expect to initially commercialize our driver out operations over the coming years. This shows the economic potential from our technology, even in the initial driver out commercial operations rollout that we expect to begin by the end of next year.
In fact, if we just zoom into one very specific route in Texas triangle, the Dallas-Houston route itself represents $ multiple billion of potential freight revenue. There are a number of industry challenges within the truck freight market that enables accelerated adoption of our solution. The first is safety. Over the last decade, there has been on average over 5,000 fatalities involving large truck crashes annually. According to NHTSA, 94% of all accidents involving trucks are due to human error, including distracted drivers, impaired drivers, fatigued drivers, or even non-compliant drivers. Our solution eliminates the human error elements involving all of those causes of crashes. The second is driver shortage. According to American Trucking Associations, the current driver shortage is 80,000 and expected to be doubled to 160,000 by 2030.
Our Level 4 autonomous trucks can alleviate this shortage and provide an increased and consistent truck capacity across the entire supply chain. The third is costs. In terms of cost structure, labor cost accounts for almost 50% per mile operating cost, while fuel accounts for about 30%. With our driver out Level 4 autonomous solution, we'll be able to lower the cost of driving with our virtual driver and also minimizes fuel cost due to improved fuel efficiency. Lastly, medium heavy duty trucks contribute 24% annual US transportation greenhouse gas emission. We have demonstrated 10% and 13% improved fuel efficiency, respectively in studies with University of California, San Diego and UPS, which will make freight transportation more environmentally friendly. From our perspective, the path towards autonomous trucking commercialization includes three sequential steps. The first is safety.
Our driver out program is our manifesto about safety. You may have heard others saying, "Yeah, we're solving all sorts of autonomy problem now, but driving for testing will be enough." We believe that the appropriate architecture for Level 4 autonomous trucking does not come from thinking about driver out. It actually comes out by doing driver out. It's only after enormous amount of real-world validation through which we simplify the overdesign and co-corroborate the underdesigns, that we can be confident that we have obtained the right solution for the Level 4 autonomous trucks. Once this is complete, we can then start optimizing the efficiency of the technology, which is the stage that we're in today. The second stage is efficiency. Since we have passed the driver out threshold, operating cost per mile becomes the most meaningful metric, which Pat will discuss in more details later.
Why do we say that this is the most meaningful metric? In autonomous trucking demonstration, you can have a good run even if there are underlying issues that limit the technology readiness for commercialization. For example, what if there's a hardware issue? What if there's an over-reliance on teleoperation? What if the map update is inefficient? This issue may not be obvious, but they will limit the commercial viability of the solution if the operating cost is not competitive in the marketplace. The third and final step is scale. After solving the safety and efficiency issues, we are ready for the scaled operation. Here, we're talking about thousands of trucks to begin with. There's no way that these trucks can be retrofitted like what we have today. They have to come off an assembly line to have low production costs and high production throughputs.
To our knowledge, we're the only company that has a concrete production contract with the OEMs, and we have been working closely with Navistar and TRATON on a detailed production plan for two years. Today, we are the clear global leader in autonomous trucking technology. We are the first and only autonomous trucking company to demonstrate fully autonomous driver authorized on open public roads. We believe that we have the most advanced Level 4 product semi-truck program in the world today. This is enabled by our long-standing global OEM partnership with TRATON and Navistar since 2020. We have the broadest and the deepest set of blue-chip partners, including both upstream and downstream partners. We are actively collaborating with our partners to integrate our autonomous technology on our Autonomous Freight Network. We have developed partnership across the entire hardware ecosystem.
At the OEM level, we have partnership with TRATON and Navistar. Across the major components of the truck, you can see that we work with world-class partners who are the best in their field for items including steering, braking, engines, tires, and sensors. One particular component I wanted to call out is the Autonomous Domain Controller or ADC. As part of our development process, we identify this as a commercial component, critical component that is not yet commercially available in the market today. For this reason, we decided to partner with our early investor and longtime partner, NVIDIA, to develop a purpose-built, high-performance ADC. This ADC will incorporate NVIDIA's powerful DRIVE Orin SoC and run our autonomous driving software. We plan to provide two service models to allow our customers to accelerate the capacity of our Autonomous Freight Network.
The first service model is the TuSimple capacity, which is active today. We are currently providing this service by operating TuSimple fleets of Level 4 trucks to haul freight for our customers. The second service model is a carrier-owned capacity. This model is designed for large shippers and carrier customers. In the U.S., customers of this service model will purchase a purpose-built Level 4 autonomous truck from Navistar, powered by TuSimple technology. They will subscribe to TuSimple Path, which charges a per mile fee for us to provide a white glove virtual driver service. While there is upfront capital expenditure, customers will be able to control their capacity scheduling, and autonomy is directly integrated into their existing supply chain. Based on the estimated savings per mile and expected extra miles driven per truck, the total cost of ownership is very attractive.
We expect the payback period will be less than one year. These two service models are complementary. They provide our customers with flexibility and also accelerate the pace of adoption. We believe our Autonomous Freight Network will accelerate the flywheel effect. Our superior user experience and reliability, differentiated capabilities, cost efficiency, and sustainability would attract more customers to use our service. With increased customer demand, we expect utilization rates of autonomous trucks to grow, naturally pushing us to add more terminals, more routes, and more truck to the network. These additional terminals, routes, and trucks will then lead to increased service capability with improved coverage, density, and efficiency. As a result, customers are incentivized to use our solution, resulting in a self-perpetuating system that further accelerates network growth. At TuSimple, we are focused on pursuing ways to create a sustainable future for both our company as well as our community.
As such, our technology and mission are highly aligned with environmental, social, and governance principles and impact. On the environmental side, sustainability is core to our commercialization strategy and product development process. Various independent studies have shown significant carbon footprint reduction of our autonomous driving system. On the social side, at TuSimple, we believe that we are not only developing cool technologies, but we're also delivering social responsibilities. We will continue to adhering the spirit to keep driving workforce diversity, to promote on-road safety, and to form an even stronger community partnership. On governance, our board of directors is directly and regularly engaged with senior leaderships and has deep experiences as top-level executives at public companies. Our board is 80% independent, 20% ethnic minority, and 40% female. In the past year, we grew our world-class team to approximately 1,400 employees globally.
We would not be where we are today as a clear leader in autonomous trucking without all of the members in our team. I'm incredibly proud of our management team. They are the reason why we are able to push boundaries and break down barriers. They bring a wealth of expertise and experience in technology, transportation, and at some of the world's foremost organizations in the industries. We believe that we have the right mix of talents to build our leading technology and accelerate our commercialization efforts. Some of our management team is here. I'm sure you will get a chance to chat with them at the reception after the presentation. Our board brings valuable industry perspective stemming from years of experience leading organizations in the technology, automotive, and regulatory spaces.
Leveraging a diverse range of backgrounds and perspectives makes us stronger and better able to strategically respond to changes in the industry. In the next section, I would like to invite our Chief Administrative and Legal-
Time out of your busy schedules to join us here in Tucson. I know you have a lot of demands on your time, and it's not easy to get to Tucson, but we greatly appreciate you guys spending your day with us here. Xiaodi talked about some of the whys. Why would you develop technology to automate trucks? Why would you have driverless trucks? The safety benefits that he talked about. Reducing highway fatalities. On average, over the last decade, been about 5,000 fatalities every year. This technology will absolutely reduce that number. It will reduce the number of serious injuries caused on our nation's highways when motorists get involved with an accident with a commercial motor vehicle. By no means am I suggesting that all of those are caused by the truck driver.
About 94% of accidents are caused by human behaviors. We think about the trucking space, think of things like distraction, impairment, fatigue. You eliminate all those factors when you have an autonomous system driving that truck. You've also experienced the truck. Has everybody been on the truck so far or in the chase van? Is it pretty much everybody? Okay. Hopefully you agree that the technology is certainly at our door if it hasn't actually crossed the threshold into the house. It's here, it's very advanced, and it's going to be prolific within the next several years, and we'll talk about that later. What I wanna talk about is how are we going about creating a scalable business model? How are we doing that?
If you're wondering, by the way, why the Chief Administrative and Legal Officer is talking to you about that, the administrative side in our world means trucking, right? In a technology company, administrative means trucking. I have responsibility and head up the Autonomous Freight Network within TuSimple. As you think about building an Autonomous Freight Network, it's not enough to just build the L4 truck. That's not enough. I have this mental propensity that I can't suppress, that sometimes I equate movies with things that are happening in my life. When I'm thinking about this presentation today, one of the things that came to me was the movie Field of Dreams. I remember where Kevin Costner was told, "Build it, and he will come. Build the baseball diamond, and they will come." Well, this isn't a similar system.
You can't just build the truck and the trucking companies and the shippers and the rail operators will come. That's not how this works. We gotta build an ecosystem. We knew very early on that we needed to build an ecosystem because this is a brand-new emerging technology, and everybody in this room knows in the freight business, it takes a long time for them to adopt change. We have to prove that we have an entire ecosystem that allows for the mass scaling of these autonomous trucks. Obviously, it starts with our leading technology and our I guess I'll advance my slides, huh? It starts with the leading technology. Because we are one of the leaders in the technology space in autonomous trucks, we have been able to create a world-class portfolio of partners.
When I use the word partners, I'm talking about strategic partners. That's my lawyer mind also telling me, "You can't say partners, strategic partners." When I say partners, that's what I mean. We have built a tremendous portfolio of world-class partners. If you look at the freight ecosystem, we're across several sectors. We're in the truckload space, we're in the parcel space, we're in the 3PL space, we're in the shipper space, and we're in the Class I railroad space. Not just with anybody, but with the country's best and most premier providers in those spaces. If you talk about the upstream partners that we have, right? Whether it's on the real estate side, we partner with Ryder on terminal space.
We just announced a partnership with Werner Enterprises on their roadside assistance program, so that folks that use our AFN system in the future, they have that roadside assistance program if they wish to use Werner's. We have property in the Alliance Fort Worth area at the Hillwood Freight Park. If you haven't seen it's enormous. You've seen similar ones, and we have a space there. We get asked a lot about the insurance aspect of autonomous trucks. How is that gonna play out in the space? We have an insurance tower just like motor carriers have. We're working very closely with our primary insurer, Liberty Mutual.
They are fully aware that this product is gonna be a safer product than the traditional auto liability space and human drivers. You also know that when they go about pricing the premiums for that, they look at your loss pick, your loss run. Obviously, with an emerging technology like this, there isn't enough data out there so that they can fully analyze what those premiums should look like. We're working with them on other things, other indicia of safety of the system. That brings me to Geotab, our telematics provider. We had an analysis done by Geotab that compared our miles and driving performance of our autonomous system versus manually driven miles of their customers.
They looked at things like harsh braking, harsh acceleration, harsh cornering, and they compare that to their customer experience in manually driven trucks. It was substantially lower instances in those categories, which, as you know, are leading indicia or indicators of accidents. It's things like that that we're pulling forward so that the entire ecosystem can make decisions such as insurance premiums. Due to this entire ecosystem, what that allows us to do is when that Navistar production-built truck comes off the assembly line that Xiaodi was talking about, when that comes off the assembly line, it's essentially plug and play because we've fully integrated with our strategic partners, freight networks, and you can just take the manually driven trucks out of the network and plug in the autonomous trucks into the network.
In 2022, we are really focusing on going deeper into our relationships with our strategic partners. I'll go through some of those examples. On the parcel side, UPS is a prime example of going deeper with our partners. We're very proud of our relationship with UPS. It's been going on for close to three years now. We have driven over 200,000 mi in autonomy with UPS. We've delivered over 6,300 loads for UPS. We have pickup and deliveries in Phoenix, Tucson, El Paso, Dallas. In California, we've taken freight to them to Orlando. We are much past a pilot or a test program with UPS. We are in fact, fully integrated into their supply chain, their freight network.
We have already started to work with them and look at the number of lanes that we can take to autonomy when that production-built truck is available. We can do it in 13 states. We've calculated the number of trucks that we think we could automate for them in their system, and it's a very, very promising relationship that we have with UPS in the future. If you look at the rail side, right? I had some discussions at lunch. We have already announced that we're going to haul freight for Union Pacific from Port Tucson, that you all visited today, to Casa Grande, which is a suburb of Phoenix. We're going to haul a container for them on a chassis. That's something that we pivoted on just recently.
We had originally thought we were going to haul it on a dry van, right? It makes more sense for them in their business model that we be hauling and able to haul a chassis and a container. The engineers and the validation is already occurring on that process so that hopefully this fall we'll be able to haul freight for Union Pacific on a chassis in a, in a container. If you look at both Canadian National Railway and UP, this allows them the ability to expand their network. We talked to, whether it's rail, whether it's 3PLs, even some truckload folks that are, you know, more in a dedicated model.
When you talk about what these L4 trucks can do, it's not just, "Hey, what can I do to put it into my current network, my current model? What other sort of opportunities can I capitalize that I can't bring into my network today?" A good example, I had a conversation with a CEO a couple of months ago, and he said, "Hey, Mullen, your technology is amazing. I don't have any use for this today, but I turn down business all the time where I could absolutely capitalize on your L4 trucks and expand my business offerings." It's not, again, just today, it's those that are progressive thinking about how can I capitalize this in the future.
Which also brings me to sometimes we get asked, "What's the point of, you know, getting into this now? You're talking about 2024 and 2025. Why don't we just wait?" My response is, and I think those that are progressive in this space is, you do that at your own risk. Think about it. If those that are progressive now and they're already building out and thinking about their network when these trucks are available, and it's plug and play for them, if you're not doing that today, you're gonna be at a disadvantage. By the way, these partners that you see here, this is not an exhaustive list. We're hauling freight for other parcel folks, for other truckload folks that are not on this list.
They just haven't got to the point where they want to be public about it. Pat Dillon, our CFO, also often talks about it as a journey for some of our strategic partners to get to the point where they're comfortable with announcing these things. We obviously honor that. If you look at the relationship that we have with Giumarra, they are a fresh produce grower and shipper, and they haul a lot of freight, ship a lot of freight cross-border from Mexico into the U.S. We partnered up with them, and we hauled watermelon from Nogales to Oklahoma City. The point of the project, so to speak, was to compare what it would look like for a solo driver to haul cross-border produce from Nogales to say, you know, interior U.S., we picked Oklahoma City.
They picked Oklahoma City. If you have a solo driver, that's about a 24-hour haul. If you do it with autonomy, where you don't have the restrictions of the hours of service and the other limitations of the human driver, it can be done in 14 hours. Which is very, very exciting to the produce folks, obviously, right? The produce gets there quicker, so there's less spoilage, and it expands markets for them, where otherwise they couldn't ship their fresh produce. The communities obviously benefit to those who couldn't otherwise get fresh produce. Now they have access to fresh produce. To some of the collateral benefits of these autonomous trucks that really three years ago you didn't really think through things like getting fresh produce to the market.
On the 3PL side, Loadsmith recently announced a 350-truck reservation with Navistar. For me, the idea and the concept of 3PLs getting involved into the asset base is extremely interesting, right? We are working with other 3PLs as well. Their CEO said, "This is an absolute opportunity for me to grow my offerings, and it helps me with my capacity and helps with the driver shortage." The 3PLs in this space I think will be very interesting as we continue to get closer to adoption and proliferation in 2024, 2025. Lastly, Werner Enterprises has been a partner of ours for a couple of years now and very progressive. We got together on this roadside assistance offering, right?
That not only can we benefit from it, they have it in their network, our other customers can use that same offering in the future. It's progressive thinking like that that these motor carrier shippers are going to get great benefit when this autonomous truck comes off the assembly line. One of the things we're really trying to pull forward are the hurdles and the obstacles as it relates to terminals, right? We're running approximately 80 trucks today, so our terminal footprint isn't all that great, but we are expanding it, and it's going to consist of our customers' locations. It'll consist of our own locations and then, of course, lease properties as well. We're currently in discussions with one of our partners on co-locating on one of their large facilities in South Texas.
It's opportunities like that will allow us to provide full service to our customers when our AFN is fully operational. If you look at the map, what we're targeting predominantly right now is of course the areas we're gonna be operating in this driverless operation that we've talked about. We'll continue to talk about throughout the day. Think of a Texas triangle, Dallas, San Antonio, Houston. Obviously, we're still running Phoenix, Tucson through El Paso. We added two facilities in Texas. Well, I'm sorry, one in Dallas, one in Fort Worth in 2021. We entered into a lease agreement with Ryder in Laredo in San Antonio. Again, we're in discussions with a strategic partner on co-locating as well down on Houston and Dallas areas.
We'll continue to focus on where our customers want us to be. And again, some will be larger terminals, some will be customer terminals, but we're also gonna concentrate on that capital light presence as well, with partners like Ryder. On the regulatory side, it's said that the environment is very supportive of the adoption and proliferation of autonomous trucks. When you think about the regulatory side, you gotta think about it in two ways. You have to think about it on the federal side and then on the state side. I'll start on the federal side. There's no prohibition on the federal level to remove the human driver from a commercial motor vehicle.
That was settled a couple of years ago, and then when the DOT announced, we'll no longer consider that the operator of a CMV has to be a human. There are some collateral issues when you take the human driver out of the truck. They are working on rules to accommodate for that. Just think of a few things. If your truck becomes disabled, you have to put triangles out, right? How are you gonna put triangles out if there's no human in the truck? How are you gonna have cargo securement? How are you gonna communicate with law enforcement? They are working on some rules that would accommodate for that. Of course, we in the industry are looking for solutions to those problems as well, right?
For instance, on the triangles, you know, we're working on developing a system where we could drop, you know, discs at the back of the truck and have them, you know, within the same coordinated 10 ft as required by this the regulation. We're thinking of ways that technology can solve these problems in the event that the regulators don't get to it. Then when you get to the state level, and of course we've talked, we're operating in the Southwest, in that Sun Belt from Florida to California. When you look at that Sun Belt, there's only one state that doesn't allow L4 deployment of commercial trucks without a human driver, and that's the state of California. The California legislature told the regulators years ago to start rulemaking to allow for L4 testing and deployment.
They did so on passenger vehicles. If you read about L4 testing in California, it's passenger vehicles. You can't test L4 in California currently. You can test L2 if you want driver assist, but that's not our space. We're in L4, so we don't test in California. We're working very closely with our competitors in California to get the regulators to start promulgating the rules to permit L4 testing. It's California, so I'm not laying any wagers, but we have made progress. I think it's kind of the consensus that when these other states start to see the tremendous benefits of autonomy that California, that they just can't really afford to not participate. They are trying to tie us to EV. Tying AV to EV right now, obviously, as you all know, just this wouldn't work, right?
The EV system isn't in place to fully commercialize autonomous trucks because the EV structure isn't there. There are some states that we continue to work on. There's some legislation going on in Kansas now as it relates to AV trucks. Oklahoma, just last week, passed legislation allowing for L4 deployment. Mississippi does not yet have a statute or regulation that permits L4 testing, but they do allow for it. It just isn't on the books yet, so we're working with them to get it on the books. The point of all that is both on the federal and the state side, the regulatory environment is very supportive of L4 deployment. Some other kind of tangential things we're working on the truck and trailer inspection process. How is that gonna look when you take the human out of the truck?
We work very closely with the Commercial Vehicle Safety Alliance, which is an association of the state agencies that regulate commercial trucks in their respective states. We've had a number of gatherings in Dallas, Seattle area. Had a week-long kind of a boot camp on what that's gonna look like last week so that we're really close on coming out. CVSA is really close with coming out those standard inspection procedures for trucks that have no human drivers. About three or four weeks ago, the PrePass folks, they're the ones that do bypass at weigh stations. They brought in about 60 people to our Tucson facility.
We gave them some demo rides and had about a half day discussion on how, again, on this PrePass, what should that look like when there's no human in the truck. We're, again, bringing all those hurdles forward, addressing those issues now so that when our purpose-built truck goes off the assembly line, we can adopt and immediately start to scale. I'll wrap it up simply by saying we firmly believe TuSimple is a leader in autonomous trucking, and we believe that for a number of reasons. Our milestones that we've hit are clear evidence that we're the first, whether it's our 1,000 m perception system, which allows us to navigate on surface streets such that we're able to do driveway to driveway, right? This isn't hub to hub. This is driveway to driveway. We do it every day.
Whether it's the driver out program or whether it's this ecosystem that we just talked about, where we're literally getting integrated today with our strategic partners and customers on how we can adopt and proliferate autonomous trucks in the network. It doesn't come easy, right? This technology is very difficult, and we obviously have world-class minds working on this technology. And others in our space do as well, right? They're all world-class engineers and minds. You look at focus and hard work, and we have been singularly focused on commercial vehicles since our inception. We work as hard as anybody. I'll tell you, Xiaodi, who just got done talking, he works harder than any, literally harder than any individual I've ever met.
Don't necessarily appreciate the texts at three in the morning, but we are singularly focused on being the first to market for L4 trucks in the system. With that, thank you very much. We're gonna take about a 10-minute break, after which, Pat and Derek and I will have a discussion and have a little Q&A. Ten-minute break, folks. Thank you.
I'm not playing around. Oh, okay. Am I supposed to introduce him? Am I introducing him? All right. Welcome back, everybody. This is really a treat for us today to have one of our longtime partners, Derek Leathers, the CEO of Werner Enterprises, here today with us. We're gonna spend the next 25 minutes or so doing a little bit of a fireside chat. As you heard Jim talk about, this is a business where we need to build the AFN with partners. They're part and parcel of everything that we do in trying to commercialize our technology. We're really honored to have Derek here. Thank you for being here with us, Derek.
Thanks for having me.
Maybe just to get started, for those of you who are in the room who maybe aren't familiar with the truckload space or haven't been familiar with Werner Enterprises, Derek, if you can just give us a little bit about your background and a little bit about Werner Enterprises, how big the operation is, where you run geographically, a little bit of background, that would be a great way to kick us off.
Sure, I'd be happy to. I've been in the industry about 35 years now. Spent the entirety of my adult career in transportation and logistics. Werner Enterprises, I've been with for 23 years, it'll be 24 in January. We're a publicly traded truckload carrier, North American focus on truckload logistics. We are about a $3 billion revenue company, roughly a little shy of that, about 8,200 trucks, 25,000 trailers. We operate across all of North America, so Canada, U.S., Mexico. A growing logistics business to augment our asset business that currently stands at roughly $750 million or so in annual revenue and growing rapidly right now.
All right. I guess as you start to think about AV trucking and how it's been evolving over the last six, seven, eight years, I'm sure you've been watching it astutely. What do you think the opportunity is to integrate AV? What are the potential benefits, and how do you think about how it might integrate with Werner more specifically as well?
Yeah, I mean, our perspective has certainly evolved over the last seven or eight years. I told the story to a couple people earlier. You know, the first few visits we made to autonomous companies, we spent more time coordinating the logistics of being able to get into and out of their building under the cover of darkness-
Mm-hmm.
-than anything else. Literally would wear hats and glasses to walk in and out of fear of how our fleet might react. I think somewhere along that journey, we came to the reality that companies like Werner have continued to operate, first off in our own network, more and more density, more and more lanes of focus, and less and less kind of all things to all people. Within that, you have a growing driver shortage problem and an aging driver demographic. It's my belief that when you look at drivers. First of all, you shouldn't stereotype any group, but drivers are no different than that.
If you were to the long-standing drivers, the people who've been in this industry a long time, they really like the open road and going out over the road, and that group is retiring in record levels. They're leaving us as quickly or quicker than at any time in our history. The drivers joining the industry are ones that really wanna be close to home. They wanna do shorter regional haul. We're working every day at Werner to try to create shorter and regional hauls and do a lot of stuff in our network today that would look like what you just saw in the presentation. An origin driver, a destination driver, and where we have enough of them, the long-haul driver that does the middle. That middle mile, I think, is an opportunity for great efficiencies over time.
I think Autonomous Freight Networks and autonomous e-technology in general is gonna play a pretty big role at some point in the future. As Jim said, we'd rather be on the front end and understand that evolution than try to play catch up later.
Right. Well, as you think about TuSimple and the partnership that's been a couple years now and continues to deepen in different ways, how do you think about working with TuSimple? What is your impression of TuSimple so far? Why did you kinda choose to work with TuSimple, and how have you seen that evolve over the last couple years?
Well, I mean, I think one thing that stands out to me about TuSimple, and it's been consistent throughout the journey, is their willingness and ability to kind of listen to input from potential partners. So whether we end up with a large scale autonomous partnership with TuSimple or not, which I believe we will, but I mean, either way, your ability and willingness to listen to inputs from us on what we view as the pragmatic implementation of something like this within our network was different with TuSimple. Like, with any startups, and especially in the arms race that autonomous finds itself in today, it feels as though a lot of companies have the rhetoric way out in front of the reality, and that presents a clear and present danger to this driver shortage issue if we're not careful.
TuSimple, by contrast, would listen and talk and want to understand, you know, what do I mean when I say that? It's fairly simple and straightforward, but we can't convince the world that there's 3 million autonomous trucks coming next week if, in fact, there is not, and wake up in a world where there's no drivers to fill the trucks that need to still be on the road. This company got that early on, and they've continued to understand that. That's just a very pragmatic example.
Whether it's a technical issue, whether it's security concerns, whether it's technology or application concerns, we've just found it to be a very receptive group to work with, and one that wants to be integrated and connected to industry leaders versus solely wanting to go out and displace that or feel as though they can go out and do it all on their own.
Yeah. I'll turn it to Jim a little bit. Jim, you've got a long experience in the trucking space. You were at Werner prior in your career. You've spent time as a regulator for the space. How do you think about sitting in your seat at as an executive at TuSimple working with Werner? How do you think about making sure that our technology is fitting into the network, and how do you kinda practically go about that process on a day-to-day basis?
Yeah. Yeah. I did work with Derek for a number of years, and it's funny, we didn't go over this together, but-
Most times in his life.
Well, if I had said my entire adult life, Derek would have said, "Well, when's that starting?" That's a whole other topic. It starts with listening, right? I do think that not only others in our space, but I think early on we might have made some of those mistakes where we thought we were gonna tell folks how they were gonna integrate autonomous trucks into their freight networks. That just doesn't work, right? Derek and his team has spent you know decades and decades and decades of figuring out how to run a successful truckload asset-based carrier. For us to go in and tell them how they're gonna use our technology is just, it's a fool's errand. We listen. Whether it's Werner or anybody else, we listen.
If you don't have trust with your strategic partner, then you add no value. You know, this industry obviously is very ROI driven. I'm not sure which would come first, the trust or the ROI or the value. You gotta have both of those, right? I think that's kind of the foundation, Pat. If you think about what's the business play, obviously moving freight for Werner is one of the first, you know, discussions we have, of course. The reservation for the Navistar truck obviously is on the table. We dig deeper and I've had folks, I hope I can say this, Derek.
I've had folks on Derek's team say, "Hey, look, we've already been told by our CEO that we're supposed to look at terminals and strategy of what our freight network's gonna look like when these autonomous trucks are available." To Derek's point, you know, will it be with us? We sure hope so. It's leaders like that that are willing to look several years down the road and kinda, again, we keep coming up pulling these hurdles forward. Whether it's Werner or UPS, or some of the others that you looked at earlier, our ability to learn from them, listen to them, and collaborate with them as to what that freight network's gonna look like in a couple of years.
Derek, you were talking a little bit earlier about some of the changes that are kind of happening in your network, whether it be driven by the driver population and their preferences. How does that sort of dovetail with some of the changes you'll you think you might need to make into your network to accommodate AV? We are, of course, trying to make this as plug and play as possible. We know that the realities of that is that there will need to be some adjustments in the way that you operate to be able to incorporate AV at large scale. Can you tell the audience a little bit about how you started to think about that for when we do have the production truck ready, how you'll. What kind of adoption or adaptations you'll have to make to incorporate it?
Yeah, I will. Obviously, we're also publicly traded, so I'll be careful with just how specific I get. At the end of the day, companies like Werner have had 65 years to build out a freight network. Take the A off the front, but a freight network, terminal network, real estate investments, thoughtful planning around how we can do more drop and hook and relay networks, how we can build more time at home for our drivers. Frankly, the good news is, a whole lot of things that we're talking about here today actually exist at large, well-capitalized carriers around the country.
Yet, despite that, the best analogy maybe I could give would be, you know, we're $2 billion on the asset side with our trucking business, and as I previously mentioned, $750 million in logistics, where other carriers do a better job. Other carriers are the better solution for certain lanes. I think the AFN is part of our network as we look forward, and we will certainly lean in to the AFN network. Clearly, a carrier company of our size believes there's size, scale, and competitive advantages to running autonomous vehicles through our own network. I think TuSimple's approach to doing both, being an and proposition is smart. It has a lot to do with why we're continuing to stay aligned, because we need it to be an and proposition. We're not gonna forfeit.
I mean, we would not forfeit our network efficiencies or our network advantages, or hand them over to a third party, but we will absolutely embrace, you know, cooperation with a third party, and we think that the network TuSimple's building out is a great example of that.
Great. Derek, shifting gears a little bit, earlier you were talking about when you first went to meet with AV companies, you were sneaking in and out. How have you managed, now that you've been public in the relationship between Werner and TuSimple for a couple years, how have you had to manage that with your driver population? What has the reaction been? You know, give us a little sense of what that was like when you first went public on the relationship with TuSimple.
Yeah, I mean, look, I've been doing this a long time, and you think over time you know all the answers, and I pride myself on being very, very close to my drivers. I spend tons of time with our drivers and in town halls and meetings and really feeling like I have a pulse of the driver that works at Werner. As it relates to autonomous, I was simply wrong. You know, we were very concerned about what that reaction might be or what the uproar might be or how much white noise might be out there. What we found is that again, the unique situation here about the timing of when this is coming along is that it aligns well with the changing demographic of the driver.
There is an old school driver out there, one that's been with us for 10, 15, 20 years, that has absolutely no interest in this or even working in this environment. That group's retiring, and they're retiring in front of our eyes every day right now as we speak. The group of drivers coming into the fleet today, the ones that are coming in through our driver academies, et cetera, they wanna be local or regional at best. They do not want to be over the road. The vast predominance of them are looking for daily home time jobs. They still wanna be a truck driver. They like the economics of being a truck driver, but they don't really want the lifestyle that comes with it, or at least traditionally came with it. There's real curiosity.
I was telling Jim earlier. I just got back from a nationwide tour of our terminals a couple of weeks ago, where I spent. I think I hit five cities in one day and four the next. At every single meeting, autonomous came up. That same question having come up two years ago would have probably caused me great fear. Really, they were just curious, they were interested, they were excited. They think it's kind of neat. Then of course, like anything, there's a subset of them that are, you know, against the whole concept and think it's, you know, devil on earth in its reality. That's the group that's retiring out soon as we go forward. Even they accept that it will happen. They just would You know, they're the same ones that didn't wanna give up a manual transmission a few years ago.
Mm-hmm.
Now happily, you know, drive automatics and understand the benefits of it.
Yeah. Jim, maybe I'll pose a related question to you. As you think about back to your time at the FMCSA, there's been a lot of rhetoric and different opinions about what AV trucking will do from an employment perspective for the driver population. What's your perspective, having been at a truckload carrier, been here at TuSimple, and been a regulator on what this does in terms of the employment dynamic for drivers, and how do folks from a regulatory perspective think about that when they're thinking about rules and regulations around this space?
Yeah. Well, I think I'll start with some of Derek's comments. Obviously he and I had these conversations about how will your driver force react to this news if you enter a partnership with us. Had similar discussions with other CEOs that we're strategic partners with. It was great to hear, it was across the board too, that they all had positive feedback from the drivers. Sure, there's a couple, you know, one-offs here and there. They're like, "Yeah," as Derek said, "it's the scourge." I don't think that you're alone, Derek, in your miscalculation. I did too as well. I thought that you'd get some pretty serious blowback. Talk about the listening aspect, right?
We're very sensitive to the fact that overselling what our technology can do and how quickly it's gonna proliferate and where does have some very serious consequences. And that funnel for new entrants to the trucking market today and the driver force is not very wide, and anything that constricts that, you're gonna cause problems. We're very sensitive to that. We're very deliberate about how we portray what our technology is gonna do. We're very specific. It'll be in the Sun Belt. It'll be quite some time before we get north. It's gonna be in high volume applications. It's not, you know, L5 go anywhere any time you want. It's not that. It's high volume lanes. We have a saying, and we absolutely believe it to be true.
If you're a truck driver today, no doubt about it, you can retire as a truck driver. I even take it much further. If I had a 10-year-old and he or she said, "I think I wanna be a truck driver," I'd have no problem. I would not steer them from that. I steer them from being a lawyer, but I wouldn't steer them from being a truck driver. I think that there is some misperception, and some of it is self-induced on what this technology can do. No doubt in my mind it's gonna help alleviate the driver shortage, as Derek referred to. I think this is still true, Derek, that long-haul driver is the hardest seat to fill in the industry, and that's our wheelhouse, right?
I think that we, the TuSimples of the world, have to do the right messaging. We can't overstate what we're up to. As far as how the government looked at it, right? The DOT commissioned a study by this organization called the Volpe Center, V-O-L-P-E. They do studies for the DOT and other analysis and other services for the DOT. They did a study, and they saw a very minimal net impact on jobs within the transportation space and also with respect to drivers. Because you are going to have probably more first and final mile drivers than you have today, which again fits our industry and the need for additional first and final and less need for the long-haul driver. We don't see it as if this is gonna be some major displacement of truck drivers in the industry.
We just don't see it that way. Lastly, let's see how many drivers we have. Probably 105-110 drivers. Their feedback is critical to not only the development, but the improvement of our autonomous system, because they are the ones that tell the artificial intelligence what is the safest thing to do under these circumstances. That's why we hire folks that have literally decades of experience, because we rely on them to provide input to our system, so we know how to react around a non-conforming motorist. Fill in the blank, 'cause there's plenty of those. There will always be that need as well. We're very sensitive to it, obviously, Pat, and we wanna work very closely to make sure that we do not do anything to further constrict that funnel into the industry.
Derek, you've been public with Werner that you work with multiple different AV technology providers today. Can you talk to us a little bit about your strategy in terms of evaluating different AV providers and how you evaluate them? What sort of criteria as you start to think about different potential partners that'll help you incorporate AV trucking into your network?
Sure. I mean, from our perspective, especially when we started this multi-pronged approach, it was still the very early innings of AV, and I think it would have been a mistake on our part to have picked an early winner before we let the technology kind of evolve a little bit and see what any one company might be able to produce and what progress they were making on their own roadmap. Secondly, there were certainly different technology paths that were being taken on how they were approaching AV, and we thought TuSimple's approach was fairly innovative. The way they were doing it was fairly disparate from what a collection of other AV companies were doing.
I'm not a tech guru, so I'm not gonna try to explain what I mean, but as my tech people briefed me, we were excited about the idea that if there was one that we felt compelled to keep a close eye on, it was the one that was doing it or approaching the problem a little bit differently. That intrigued us early on with TuSimple. It continues to intrigue us today. Yes, we do continue to work with and, just like you work with other truckers, we work with other AV companies, and over time, the cream will rise to the top.
Yeah. Jim, how do you think about the customer relationship across the different fleet customers that we have here? You know, as you've kind of evolved in this job, how do you think about what are the different elements that help drive customer success and customer satisfaction, both at this stage, where we're still relatively early stage, and what's gonna allow us to be successful as we move into the middle part of this decade and start to scale up?
I wanna circle back to something that Derek just said real quickly, you know, they're working with other providers. We deal with other trucking companies. There's one of the country's largest truckload carrier that we met with their executive team might have been a year ago, right? Literally they said, "Don't see it. Don't see how this is gonna work. Don't see how it's gonna be a business. Don't see how it's gonna be more productive. Don't know how it's gonna fit in the network." Okay, all right. Within eight months, they announced a partnership with a competitor. Somebody asked me, "Aren't you mad?" I'm like, "No, I'm not. Don't care." What was important to me was one of the country's largest truckload carrier just said, "This can work.
This can work." What you just said was the cream will rise to the top. Look, we think that we are, you know, the cream, and so we think we're going to be the leader in this space. I'm not overly bogged down on, you know, what we can't have everybody on board at this point in time. What I think adds value, Pat, is again, the listening, working very closely and spending the time to learn their network and how our system can help, where there's greater capacity, where there's more reliability, whether it's the cost structure, whether it's the safety component. How can we help you become a better motor carrier? If you can't prove that, if you can't show it to them.
That's why we often get asked. We talk. Xiaodi showed the TuSimple capacity and the carrier capacity. We often get asked, you know, "So how big of a fleet are you gonna be and what's going on on that side?" We think it's important that we be the first adopter of this technology. We don't think that you can hand an R&D project and that you've got maybe to commercialization in theory and say, "Here you go, Werner. You know, you go figure out how you're gonna implement this in your freight network.
Here, by the way, pay us a bunch of money." The way we look at it is we have to integrate completely with our strategic partners to make sure that we understand their business, that we can actually add value to their network, and that time when the Navistar truck comes off the assembly line, it's plug and play, and we're already fully integrated.
All right. Well, Jim, Derek, thank you guys very much for this discussion. And Derek, in particular, thank you for being here and supporting us. We value your partnership. We'll take a quick break, and then we'll come back with the rest of the presentation.
Thank you.
Thank you.
Hey, welcome back, everybody. Next, we are gonna talk about our Asia Pacific business, or what we call APAC. As you've seen earlier this spring, we did announce that we are gonna be exploring a potential subsidiary transaction for the business. The process is still at a very preliminary stage. We are considering a range of potential transactions, including a sale of a minority or a majority equity stake in the business. One update to our prior statements on this, we are now also gonna consider a complete sale of this business. As we have stated previously, we think this business has tremendous intrinsic value. We think that intrinsic value is not currently reflected in our share price. We are gonna explore a range of potential transaction structures that we think could unlock maximum value.
It is really too early for us to speculate about the potential transaction, but as soon as we have any appropriate updates to share, we will certainly provide that to the market. With that little bit of preamble, we'll dive in and provide a little bit of an overview on our operations in Asia. First, as you can see from the numbers here, it is a pretty substantial business. We've got over 400 R&D personnel in China. That's split between a research center in Beijing and testing operations in Shanghai, which I'll talk about in a little bit. From an IP perspective, we have over 270 patents issued in China.
We have major operations that include the testing operation at the Shanghai Deep Water Port, and then we also have other efforts in broader Asia, including a development program that we're doing with a major OEM in Japan. To date, our Asia Pacific operations have generated over 500,000 real-world road mi in testing. We have approximately 25 trucks that are running L4 technology in our fleet, and they're all equipped with the same proprietary 1,000 m perception technology that we have here in the U.S. We also have a network of world-class partners here. That includes NVIDIA, who is working with us globally on the Autonomous Domain Controller. It includes other global partners that includes ZF from a supplier base and then TRATON, given that it's a global OEM relationship with them as well.
From an OEM perspective, we also have a relationship and work with Foton, which in China is one of the major OEMs in the region. Very full suite of partnerships similar to what we have globally, also present in China. A little bit about the go-to-market strategy in China. We always talk about this. It is the second biggest freight market in the world, so it's a massive economy. It's a massive opportunity for us. The growth that's projected in that market is also similarly quite impressive, with a 6% CAGR expected to grow to $1.3 trillion by 2030. A massive market, a huge opportunity for us. To serve this market and to commercialize our technology, we're gonna have two distinct offerings.
First, similar to what we're doing in the U.S., we are gonna be building an L4 Autonomous Freight Network in China, and that's to bring driver out autonomy to the most dense freight routes in the area. We expect to go to market similar to the U.S. both with carrier-owned capacity and with TuSimple capacity. Closely mirroring what we're doing in the United States. Right now, full driver out testing and operation of an L4 commercial vehicle in China is not allowed. Just given that regulatory uncertainty, we are also gonna develop a second prong of our strategy, which is building an L2 driver in ADAS solution for the Asia market. We're doing that in large part in collaboration with our partners at NVIDIA. It's building off of the partnership that we've announced with them around building the ADC.
We're planning to offer an integrated hardware and software L2 solution for the Asia freight market, and we're also exploring on a preliminary basis some opportunities to expand that L2 solution into passenger vehicles as well. I would just say, please note that because we are pursuing a potential L2 product in China, our U.S. team here is fully focused on L4 only. Really this is just a product of the regulatory environment in China, and our entire U.S. team is focused entirely on a driver out L4 solution here in the U.S. as well as in Europe. A little bit about the commercialization roadmap in China. First, we're, as I've mentioned before, intending to offer the L2 driver in solution by 2023 as a full stack solution.
We expect to market the solution both to OEMs as well as to Tier 1 suppliers. We think this type of approach is gonna allow us to get to market more quickly. It benefits from our strong know-how in hardware selection, in architecture design, takes advantage of our deep algorithm and development capabilities and our broad expertise in various types of different sensors and other hardware. We expect that this L2 development roadmap is gonna be done in a way that will avoid undo bandwidth drain on our L4 development, which is gonna continue in parallel. We're trying to be very conscious of any kind of bandwidth drain as we're developing both of these models in the region.
On the L4 product, we will commercialize in Shanghai first, specifically at the Shanghai Deep Water Port, and then we'll expand to other similar infrastructure type projects, as well as systematically expanding to other roadways and use cases. It is a little bit different use case, at least in initial commercialization relative to what we're doing in the United States. From 2025 onward, we plan to expand the AFN regionally in kind of key freight regions, and then eventually building up to a national AFN. A little bit about the Shanghai Deep Water Port project, which you've heard us talk about previously, and I mentioned in a few slides earlier. We think this is really a sign of some considerable progress that our team has made. A little bit about the operations.
Just like what you've seen in ports around the world, efficient operations, particularly moving freight on and off seagoing vessels, is more critical than ever in keeping supply chains moving smoothly. SIPG is the operator at the Shanghai port, which by the way, the Shanghai port has the distinction of being the world's busiest container port. It has over 40 million TEUs that are passing through its operations annually. The Shanghai port is connected to Shanghai's Pudong New Area by the Donghai Bridge. This is a 20 mi bridge that typically has 20,000 truck trips per day, so it's an incredibly busy, dense freight lane.
Because of this increasing global truck freight traffic that goes on this bridge, and in the port operations, SIPG is looking to incorporate autonomous trucks into its operations, and that's both for a safety benefit as well as an efficiency benefit. It's a very clear use case for Level 4 autonomy to really improve the operations within that port. I would also say this type of port environment is incredibly complex, especially with the bridge element. It requires very complex local route and driving behaviors, navigation of frequent road maintenance operations, heavy traffic congestion in peak hours. Because we also do reverse drive parking maneuvers in the port, it requires a level of accuracy up to 10 cm.
It's a very complex, technically difficult operation, and we think it's a really good proof point around the strength of our technical team and what they've been able to accomplish in the region. As I mentioned before, we've accumulated over 500,000 real world mi, primarily at this port operation, and demonstrating really significant capabilities. This is all backed up by the fact that we are the only non-state owned enterprise AV company that is able to operate at this port, which is a national strategic asset in China. We think it speaks to the strength of not only our technology, but our presence and our commercial relationships in the region. I'll sum up a little bit about this very brief overview of our China business.
We think it's incredibly well positioned to commercialize a very large truck freight market that has a really compelling need for this type of technology. Our presence in Asia goes all the way back to our founding in 2015. We have a long-standing presence there. We have a world-class team of artificial intelligence experts, hardware experts, and folks that know what it takes to be able to commercialize an L4 truck in the region. And our technology, we think is second to none in the region. With that, we'll hand it off to, I think Xiaodi is gonna come up here next, but wanted to provide a quick overview on the China business. I'll bring Xiaodi up to the stage to begin our technical deep dive.
[audio distortion]
No, it's preliminary right now, but we expect, I think we've made the comment that we do expect to hope to have a resolution by the end of this year.
Okay. Thank you, Pat. Hello again, everyone. In this section, I'm gonna talk about our technology development of TuSimple.
First, I wanted to talk about something about the general trend of technology. The technology development of autonomy is a massive engineering project. Like every major engineering program, whether this is to develop a chip, a rocket, or an autonomous vehicle, there's always a certain natural sequences that we must follow. In order to be qualified to discuss the commercialization of the technology, we must first demonstrate the driver out capability. Otherwise, it's always assisted driving. In order to launch the vehicle for a driver out run, we must design a thorough validation program so that we can understand, every conceivable type of failures of the system. We have to prove by data that our system can always actually guarantee safety. In order to pass the validation, redundancy for both software and hardware must be designed and implemented.
Needless to say, all of the aforementioned categories are built on a complete set of Level 4 autonomous driving features. On the left-hand side, you can see there's a tree structure down in there. First, we can talk about the features. Features are easy to understand. In this presentation, you will see that feature development is far from the whole story. Even if we have done driver out, it doesn't mean that TuSimple won't develop new features, because feature development is actually an endless process. We always wanted to challenge and exceed the current limits of human driving capabilities. It comes to redundancies. Redundancy is an understated topic in autonomous driving. Many people demonstrate redundancy by just showing extra sensors. The reality is far more complex than that. It goes beyond just hardware, but includes a full detection and mitigation plan of any failed hardware or software.
Demo run is not the best form to demonstrate redundancy because 99% of the time you do not encounter any scenarios that requires redundancy. In a driver in demo, the human driver becomes the ultimate redundancy that covers up the functions of many software or hardware components. It creates an illusion that assembly of the feature is enough to deliver zero disengagement demo. However, in a driver out scenario, the formula changes. We must complete all of the redundancies from design to implementation to validation before we can allow the vehicle to go out without a driver. When it comes to validation, there's a disproportionate emphasis on simulated miles or road test miles. However, we have to remember this, validation is the purpose. Simulation or road tests are means. As Murphy's Law states, all components of a truck can fail, and they will fail.
The validation process to hypothesize every possible failure and then build a mitigation plan so that the system remains reliable even if the component fails. The validation process cover all failure signals from the engine, steering, braking, power, in-vehicle network, sensor, server, operating system, and algorithms. Traditional simulation only examine the algorithm aspect of the system, but in the following presentations, you will learn how we design our validation to cover everything from software to hardware. Next is that we have heard other players talking about our driver out programs. They will say that, "Yeah, we can do exactly the same thing, but we just have a driver in for precaution." The hardest part in autonomy, I tell you, is to eliminate that tiny portion of reliance of a human driver. It's all or none game.
Until you have designed, implemented, and validated all of the mitigation plans, you simply cannot remove the driver. There's no such thing as driver out until you have actually removed the driver. We have, on several occasions, mentioned about our operating cost of autonomy, because there is a widespread misconception that once Level 4 autonomy is achieved, there are no ongoing operating costs for the virtual driver. This is not true. Obviously, autonomous driving requires additional sensors and computing units, but what have been overlooked is the recurring cost of operating and maintaining the autonomy suite. The sensors need careful maintenance. The replacement of failed hardware requires highly skilled engineers. Furthermore, map updates, teleoperation, other IT costs also need to be optimized. TuSimple's goal for the next two years is to keep driving down the operating cost of the autonomous driving fleet.
The inflection point of this process is when the operation of autonomous get cheaper than that of a human driver. It is not the endpoint, but having a competitive operating cost is a key waypoint in the path to wide-scale adoption. To get to the inflection point quicker, we have mobilized a company-wide project called Penny Saver to seize every opportunity to reduce the autonomy cost. In the following presentations, we will show you the analysis of first batch of our driver out runs. The efficiency optimization framework will serve us as a series of milestones for us to hit in the next several years. All right. In this section, I will introduce TuSimple's advanced planning features.
Here is my favorite quote from the famous astronaut, Frank Borman, "A superior pilot uses his superior judgment to avoid situations which requires the use of his superior skills." This is the goal that we strive to achieve for motion planning. TuSimple's design philosophy for planning is like this: behind one disengagement, there are 10 near misses and perhaps 100 imperfect handlings. Avoiding tricky cases is more important than handling the tricky cases. Our design focus has advanced from maneuvers to mini maneuvers. What are the mini maneuvers? In the demo ride, you may have experienced a subtle speed change or a small lateral shift. Sometimes you will not even notice if you don't look into the HMI. These nuances in vehicle driving can change the state of the interaction without qualitatively changing the state of the vehicle.
In a ride, we make a lot of these mini maneuvers just to avoid dramatic maneuvers. The richness of mini maneuvers, one of the signature of the TuSimple. Let me give you three examples. The first one is the safety shift. In this scenario, an adjacent vehicle begins to invade into our lane. Even though it's the other vehicle that is driving erratically, we can still take proactive action to improve the safety for everyone on the road. When they get too close, we very precisely shift over in one lane in order to give room for the other vehicle. This is especially important when the adjacent vehicle is a large truck, because there's very little room between these two vehicles. We believe we are the only autonomous trucking company that has developed such ability to manage the scenarios. Why this technology is challenging?
In short, we have to find the right timing, which relies on our superior perception and braking for the right amount, which relies on our superior control. Let's take a look. You can see this truck passes us on the left side, has moved dangerously close into our lane. With only a couple of meters between the wide bodies of these two trucks, executing a safety shift keeps a safety distance between the truck. Okay, we encounter a vehicle invading into our lane on nearly every single drive. That's a very common situation that we deal with every single day. Here are just a few more examples why having the safety feature is so important. I believe most of you have experienced that on the road already. The next case study is about handling aggressive cut-in by other vehicles.
Because of the size and weight, large trucks typically take longer to accelerate and to decelerate. This means that there is often spaces in front of a big truck that a smaller car likes to cut into as they pass the truck. That often creates an unsafe distance between our truck and the vehicle ahead. Why this maneuver is very hard? To simply avoid collision, that's easy. Just brake whenever there's a cut-in. However, it's extremely uncomfortable and also not safe to the vehicle behind us. To be comfortable is also simple, just not to brake too much. However, it is extremely dangerous and has a high risk of collision with a cut-in vehicle. The hard part is to make the right decision based on the real situations.
Similar to safety shift, to make the right brake decision, the requirements of perception and control accuracy are both very high. Let's take a look at an example of a driver out run. Pay attention to this white vehicle on our left-hand side. The vehicle cut in front of us. Note that we have reduced our speed about 3 mi per hour. The passing vehicle is traveling at a brisk 69 mi per hour and has a clear space ahead so that it will continue its fast speed. Taking these factors into account, our algorithm knows that reducing our speed by only about 3 mi per hour will create a safety distance between our truck and the passing car and avoid having a harsh brake event. Now, let's watch some more dramatic examples. On the top example, we have a big truck cutting in. This is very dangerous.
In the bottom example, the pickup truck also changing two lanes on the right. Note that in both of these cases, we do not have a foundation brake. Okay. The third case study is about proactive changing lanes and handling merging traffic. Dense traffic lane changing is a headache for everyone. For a truck that is three to four times longer than a sedan, this is a much bigger challenge. In order to perform high-priority lane changing, such as getting off the highway, our truck proactively seeks the right window and negotiates with other road participants by adjusting our speed to catch that window. Why is it hard? When the traffic is dense, it takes a very good perception and prediction model to adjust the speed, location, and most importantly, the intention of all of the road participants.
We will focus on the most appropriate opening in traffic to merge into. Sometimes this opening in the traffic is ahead of us, sometimes behind us. Sometimes the gap doesn't exist yet, but our prediction model tells us a likely gap is being developed. In a word, we have to adapt to a highly interactive decision-making algorithm to accomplish the negotiation. Let's watch an example in our driver out run. We are making lane change in order to make room for the another truck that is merging in onto the highway. Our system predicts that this vehicle will try to merge in directly into our vehicle lane. Unless we move to the other adjacent lane, a foundation brake is unavoidable. Well, technically, we still have the right of the way, but taking this proactive lane change will significantly increase the safety for all road users.
After making this merge, we go to the right lane for the courtesy on the road. Again, just like the other cases, we encounter this situation all the time. Here are a few highly complex scenarios that our truck is encountering. Noting one example, our autonomous system identifies an opening in traffic behind it, and we slow down for that gap. A second example shows our truck simultaneously negotiating multiple safety shifts while navigating to a safety lane change. Okay, let's change gear and discuss about our advanced control capabilities. The goal for control, as always, is to surpass human-level maneuvering. It is obvious that driving a truck is full of unique challenges that are different than a typical passenger car. Its wide body means that there's only 2 ft on either side of the truck when it's centered in the lane.
It has 12 gears, and it takes three seconds to shift between each one, whereas cars take only up to 0.5 second. The trailer weight varies, and the distribution of the weight within the trailer is often unknown. How about making a turn? Intersection turns with a 70 ft-long vehicle can be challenging, even for seasoned professional drivers. All of these factors affect how we design precise vehicle control for the autonomous truck. In TuSimple, we have pushed the performance of our control system beyond the level of an average human truck driver in several ways. For simplicity, there are two easy indicators that everyone can ask every autonomous driving company about. One, lane centering. This is about the lateral control precision. Our solution is 2.4 times better than a human driver at staying center in the lane.
I believe some of you have experienced that already in today's ride. Second is about stopping accuracy. This is about the longitudinal control precision. The stopping accuracy of our vehicle, even on a possible uneven slope, is within eight inches. It is also worth mentioning that all of the data are measured with the trailer with cargo. How does TuSimple achieve such precise control? The control-by-wire interface or the control interface is a protocol by which the computer sends a signal to the actuation unit, to throttle, to braking, to steering. In the early days of TuSimple, this interface are inherited from the Level 2 adaptive cruise control functionalities. Basically, you dial in the target speed, the chassis does the rest for you.
It's simple to use, but you cannot do very complex action because you're only allowed to tweak the cruise control speed dial. To increase the capability to control the vehicle in a more fine-grained level, TuSimple has invested years' worth of co-engineering in defining the advanced interface with tier-one suppliers. We finally reached a level of completion in hardware interfaces for braking, steering, and drivetrain in 2021 that fulfills the requirements for driver-out operation. From the above chart, you can see that there are lots of gold stars, all of which represent that TuSimple's world's first pioneering work in collaboration with tier-one suppliers has become an industry standard. Now we're working on defining the next set of interfaces with tier-ones, including precise real-time diagnostic for steering and cooperative redundant braking.
These are the advanced capabilities that continue to improve the safety and efficiency that should be ready for deployment in about 2023. We believe we are the only autonomous trucking company at this level stage of development on hardware and interfaces. Let's dive deeper into a case study to our work with Cummins on the drivetrain. As I mentioned, all autonomous driving starts with a level two assisted driving interface. You dial a speed, the system does the rest for you. With this interface, you can't perform any high-precision control. There are other intermediate interfaces that other companies may have used, such as a synthetic pedal position input or acceleration demand for the whole vehicle. These interfaces are designed with either level two scope with a human driver present or with very limited ODD, operational design domain, where the scenario variation does not match TuSimple's commercial needs.
Now, with the Cummins, we can directly control the engine torque. This is like driving a Formula One car. You can run very fast, but first you have to be extremely skilled at driving. With the deep collaboration with Cummins, we're also working together on other advanced features such as reversal, neutral gear control, and combustion engine level fine-grain fuel optimization. In addition to the deep interface collaboration, another secret sauce of TuSimple's control system is our algorithm breakthrough. To our knowledge, almost all of today's autonomous trucking companies uses the passive controller, which is mainly based on the loop of observing and correcting the system state, like shown in this illustration.
Thinking of an experienced human driver driving in their favorite car on a road they know well, they will develop an intuition about how the vehicle handles, how it will respond, and what it is capable of. This is exactly the spirit of TuSimple's predictive control system. We model all the vehicle dynamics and deeply understand the control performance of the truck. First, we have to understand the system deeply, down to every variable, torque, accelerating dynamics, air brake lag dynamics, gear transition timing, and many other factors. Then we build that knowledge into the motion planning algorithms. In this way, our AI system can develop a very similar intuition for how the truck drives. We fine-tune our driving to predict the truck's control error and tighten the truck handling. This makes a big improvement in safety and fuel efficiency.
Let's just take one very reliable, related example. Because of the size and weight of the Class 8 truck, driving on a hilly road has big impact on fuel economy. The driving behavior and vehicle control on these hills are huge sources of fuel savings. The driving behaviors and control is designed to capture momentum on the downhill to use in the next hill climb. We allow the speed to increase slightly above our target while still operating under safety, and the speed limits. On the uphill, we allow the speed to drop below our target as we approach the crest of the hill, since we know that we can easily gain that speed back on the next downhill. While the OEM and the engine manufacturers control fuel injection, aerodynamics, and other design decisions, only the TuSimple's driving company can use driving behavior and control to save fuel.
This is a key element of TuSimple's strategy to provide a value through fuel efficiency to our customers and reduce greenhouse gas emission. Next, we're gonna talk about our advanced perception. At TuSimple, we have a rich lineup of high-performance sensors, cameras, LiDARs, and radars. Back in 2019, TuSimple had 100% camera-based Level 4 solution that could achieve hub-to-hub operation. Over time, we have added radars, LiDARs, and long-range LiDARs. Why? Because focusing on one single physical mechanism to sense the world is not robust enough. Not only do we need to answer questions about hardware failures of the system, but also possible bad or even extreme weathers. Balancing use of the sensors, be prepared for the failures of any physical measurements, is our solution to robust perception.
To leverage our powerful sensor suite, including camera, LiDAR, and radars, we develop deep learning and machine learning algorithms to help the truck to understand the world. Very importantly, our perception stack is not only powerful, but also robust. Let's see how we do it. TuSimple's perception stack is composed of layers of features: pixel level, object level, and scene level. Like you see here, pixel level features provides faithful representation of the world, just to name a few. Image segmentation tells us the object category of each individual pixel, such as road, vehicle, or vegetation. Semantic rendering connects our observation with map information, such as lane ID or road curvature. Optical flow provides dense correspondence between nearby frames, which helps us understand the motion of both moving object and stationary objects.
Lastly, occupancy grid infers fine-grained occupancy and occlusion information in the 3D world, which is critical to safety. These imaging techniques help us build a fine-tuned and robust view of the world around our truck. The next level is object-level representation. Our state-of-the-art 2D and 3D object detection, fusion, tracking, and state estimation algorithms accurately captures the physical properties of the other road participants, including their size, class, location, orientation, velocity, turn signal status, and potential interaction with our truck. In this example, we recognize that the braking and turn signals of the other cars ahead of us, which help us smoothly navigate when we approach the traffic jam. I believe we do have that traffic jam earlier today in one of the demo runs.
Other example of this smooth interaction include conducting a safety shift for nearby semi truck or canceling a lane change for fast-approaching cars. In addition to detecting and identifying objects and vehicles, we also conduct scene level perception. At a scene level perception, the scene is formed by a lot of individual objects. By understanding many objects together in context, we identify the specific traffic scenarios. Here's an example of the construction zone handling, which has signs, cones, and other vehicles. Let me repeat that video again. Okay. Yeah. Notice our truck identifies the merging to the left signal and making the left lane change, identifies the area to avoid with cones and properly handling the traffic within the construction zone. Okay. Let's look at a perception case study for how can we safely interact with pedestrian along the highway. First, let's see how we handle pedestrian during daytime.
As some of you have seen in a demo run today, our system could detect and react to pedestrian running across highway promptly, safely, and smoothly, thanks to our camera perception module. However, what if it happens at night? The visibility range of the camera is greatly impaired during nights, so relying on camera detection alone is not a good idea. LiDAR detection naturally becomes the backup solution. However, any machine learning algorithm could not guarantee 100% detection. How could we ensure safety? Independent to the object detection approaches, we design another perception module called unknown object detection to deal with these type of scenarios. This module helps our system to understand every LiDAR points using prior knowledge from the map, localization and detection. For example, is this point from the ground, the tree, the vehicle, or something else?
This module outputs objects formed by unexplained points, including the pedestrian walking along the highway at night. Yeah, let's watch that video again. Together with our powerful state estimation and tracking modules, which stabilizes trajectory and handle occlusion, our truck sees the pedestrian at about 150 m, well before humans. Finally, our planning module makes conservative decisions for such human-like object by safely performing lane changes. Unknown object module is not the end of our safety frontier. There is a chance that due to system error, for example, the server has some problem, we would go into the MRC state, minimal risk condition, and our perception capability is impacted. We build another parallel module called occupancy grid, to ensure that we never hit objects that occupies the road. As you can see, our multilayer perception solution endows our system with substantive safety robustly.
These techniques applied across the board to all types of objects and road users. Here, instead of a pedestrian, we see an example of a car stopped on the side of the road at night with lights off. The vehicle could be very difficult to see by the human eye, but our solution correctly identifies the car, and we shift over the lane to give some space. Next, I would like to introduce our VP of Hardware, Vince Verna, to talk about redundancy and hardware ruggedization.
Thank you, Xiaodi. Good afternoon and welcome. My name is Vince Verna. I'm the Vice President of Hardware at TuSimple. You know, I started off my career at traditional automotive companies like Ford, Lear, and Mercedes, where I learned the fundamentals of robust design. I did that for about 20 years, but most recently, I've had the opportunity to work at startups like BYTON and Zoox and now TuSimple, where I've had the pleasure to work with amazing people in cutting-edge autonomous technology. Today, I'm gonna talk to you about hardware ruggedization, why it's important for driver out, and what TuSimple is doing to make OEM trucks ready for autonomous operations. It's just the fact that standard OEM truck platform hardware has failure modes that are unacceptable for driver out platforms. What does that mean? I'll give you some broad categories of examples.
Cameras, LiDARs, radars, compute units, et cetera, they have less inherent reliability than standard hardware because it is often cutting-edge technology with less years of testing and less road time than standard components. What can happen when autonomy hardware fails? When sensors or compute units fail, they can lead to wrong decisions being made by the autonomous system. With chassis components like steering and braking systems, there may be single point failures which would be acceptable with a driver, but not acceptable without a driver. For example, if the steering or braking systems suffers a failure, you could still control those systems manually, but with some difficulty. This is impossible to do with an autonomous system. We need to ruggedize it by adding redundancy. Finally, on the electrical side, what does an electrical touch? Wiring carries power, networking, and signals throughout the vehicle.
That means we need to foolproof every single wire in the vehicle for proper routing and protection. We need to make sure that the vehicle's power system can survive a catastrophic failure, like the loss of a main power cable. Again, a standard truck platform does not have any mitigation for these failures because they're usually not needed since the driver is the backup system. How does TuSimple improve the standard OEM truck platform? First, we need to talk about the roles. There are OEMs, there's Tier 1s, and then there's TuSimple. Their traditional relationship for developing new technology would be the OEM working with a Tier 1 supplier.
TuSimple sits somewhere in the middle, helping the OEMs to add the hooks required to install an autonomous system, and we work with Tier 1 suppliers on the other side to understand how they need to redesign their components for ruggedization and redundancy. TuSimple has made significant investments in its hardware team to bolster and support our cutting-edge software. For example, our mechanical team takes the OEM platform and through 3D modeling and FEA analysis, ensures a perfect fit, robustness to vibration, cooling of the computing resources, and protection of our sensors. Our electrical team designs in-house electronics that are critical to supporting the autonomous system, electronics that are not available off-the-shelf.
Our testing team puts our autonomy through rigorous testing based on automotive and commercial vehicle standards. With this significant hardware capability in mind, I would like to take you through two case studies that showcase how we ruggedize a standard truck platform. The first case study involves the Controller Area Network or CAN network. This is like the networks we all have at home. It's just in your car. What is shown here in red is an interruption in the CAN network. This message is carrying important information to the brake controller, but this message can be interrupted if the wiring suffers a failure, like a short circuit, a broken wire, or corrupted data. A failure to send this message can cause the vehicle to execute what we call a minimal risk condition maneuver. It then uses a redundant CAN bus to bring the vehicle to a safe stop.
This is okay, but not ideal as the mission needs to be canceled. How do we protect these important CAN wires from this sort of damage? Well, there was nothing available on the market, so TuSimple designed and built its own solution for protecting the CAN wires. It's called the CAN bus distribution board. Instead of having all the CAN wire connections being made throughout the vehicle where they can be pinched, cut, or otherwise damaged, we make all the connections right on this printed circuit board. This increases the robustness of those connections since they are not exposed to vibration or chafing against other components, and we are protected for the future. You see all those gray connectors? Those spare gray connectors allow us to be able to add future functionality without adding more wiring to the vehicle. We can just plug them right into those connectors.
The next case study is power ruggedization. This is one of the biggest failure modes in autonomy. In standard OEM truck power systems are not built to the level of safety and autonomy required for an autonomous system. They only have one source of power. There is no redundancy, and there's no way to handle an open or short, except to shut down that system. Normally, that shutdown would happen in a very uncontrolled manner. TuSimple's solution is our in-house design power electronics. One part of it is called the PBR or Power Bridge, and the other part is the power distribution unit. Together, they function as an ultra-fast safety switch. What they do is they isolate the fault, they reroute the power automatically to a secondary power source, keeping the autonomous system up and running.
Again, these electronics were not available off the shelf, so we had to design them and build it ourselves. Having gone through these two examples of ruggedization improvements, you may ask, how do we know that these designs are ready to be put on a truck for years of reliable service? We put all of our parts through our hardware validation process. In hardware validation, components must pass rigorous automotive and commercial vehicle standards before they are installed onto a truck. In addition, they must pass several levels of testing through what we call the V-D evelopment process. This increases our confidence in the component at every step before it is put on the truck. I will show you a detailed example with our power bridge module.
The power bridge module is an internally developed component, which I introduced earlier when I talked about adding robustness to the truck's power system. You can see here that first we test the unit on its own. We test it at hot and cold temperatures. We vibrate it. We simulate worst case road conditions. We submerge it in 1 m of water. We stress it in every way possible to ensure that when we put the PBR in a larger system or on a truck, that any failure would not be attributable to the PBR itself. After the component testing, we put it into subsystems testing. This is testing where the PBR is connected to everything that it would be connected to in the vehicle power system.
We do this to make sure that the power system is 100% functional before going to the next level of testing, which would include the entire autonomous system. Next, in systems testing, we put the PBR again into a full autonomous system now, knowing that it has passed the lower level test before it. This allows us to focus on testing the autonomous system without worrying that the power will go out and with confidence that there's gonna be steady and clean power being provided. Here we can inject test vectors for all the serious failure conditions, including sensors, compute hardware, wiring, and software. This is the last step before it goes onto the truck. Then finally, in truck testing, we install the PBR onto 21 trucks. We gathered a lot of data while they were completing missions.
Over 18 months, we put 750,000 mi of testing on the PBR. During that time, we surfaced 77 issues which were resolved, leading to design improvements in hardware and software. Now, all of this ruggedization and testing is great, but if we can't commercialize our product at a cost that is lower than a vehicle with a driver, then none of this matters. Many times when we solve a problem, especially in the hardware area, it directly leads to a cost per mile reduction. Here are two examples of issues found during vehicle testing and how we take issues to our root cause analysis process, track it down to the exact cost per mile contribution. In example one, we had a CAN network communication error, which had caused disengagements on a few trucks.
We tracked it down to a hardware failure with a part that had the wrong value. We redesigned it, and this resulted in no more disengagements being caused by this issue and a 2% reduction in cost per mile. When we calculate cost per mile, we include truck downtime, parts, labor, engineering, and testing costs. In example two, we had a failure in the arms that hold our LiDARs. This could result in the sensors coming loose, misalignment of the sensors. The root cause was a supplier quality issue with incomplete weld penetration. We worked with the supplier to create a better process to solve the issue, and we were able to attribute 1% cost per mile to this issue. Those were only two examples.
We triage hundreds of issues which turn into dozens of projects that we run every quarter, leading to continuous improvements in our hardware reliability as well as lowering our cost per mile. You know, most people think of TuSimple as a software company, but I've taken you through our significant capability in hardware, ruggedization from design to testing and validation, and problem-solving that eventually leads to not only robust hardware, but cost reductions that put us on the road to commercialization. We provide all these lessons learned to our OEM partners, resulting in a significant competitive advantage. Thank you. Next, I would like to invite Ersin Yumer, our Vice President of Algorithm, up to the stage.
Thanks, Vince. A little bit of a background on those that might not know me yet. I have been at Argo AI, Uber ATG after that, and Aurora before joining TuSimple. I have been around the block for a while, leading various teams from simulation to research to engineering and operations. Here I help with leading the algorithm team as well as the test ops team. I think it's also fair to say that I'm probably the nerdiest of the bunch that you'll be hearing from today. The jacket I'm wearing is not necessarily by choice, but more of a gesture to our beloved CFO, Pat, so that he's happy. With that, I'm going to jump into starting with asking you a question.
What do you think this number represents? Just think about it for a couple seconds. This number actually represents our ADS, the miles that our ADS drives in simulation for every one road test mile that it drives on the road. If you think about this number, I'm pretty sure some of you are impressed by it, some of you are not impressed, and some of you are unsure what to do with it. The really interesting part of this is, with all the players in the industry, you are being thrown at different numbers about how big and how small their simulation miles are. How should you differentiate between these numbers that are basically thrown at you? Well, I'll answer that question pretty soon. First, let's watch a truck drive.
What do you think is interesting about this particular driving scene? Well, as most of you are already guessing, nothing really is, right? It's an empty, straight road with minimal driving skills required to essentially proceed. Why am I showing you this? The reason is essentially that there are many, many miles that are driven by these vehicles, especially on the highway, that are basically boring, right? With no dynamic objects, interactions, and driving on pretty much a straight line. Most highway driving is like this. Whereas there are interesting miles that every now and then happen, where an interesting interaction or decision leads to a precise maneuver requirement, such as accepting a merge from a ramp.
A blanket statement such as, "We drive millions of miles in simulation" is not really a convincing argument. The content of those simulation miles, the information such as the scene and what it contains is actually what matters. This is what we focus on at TuSimple. This is important to think about when you think about millions of miles or billions of miles that are driven in simulation. What is the quality of those miles? I wanna go back to this number and answer that question that I punted on, which was how do you believe the necessity or sufficiency of just many simulation miles? The answer is that you shouldn't. It's not only the number, right?
It's the richness as well as the coverage of the content that's inside is what is most important. If I'm boasting to you about just the number, you should immediately think that I'm either I don't know what I'm talking about or I'm simply trying to refocus your attention to a number that I can easily quantify, but not necessarily qualify for. With that, we talked about the richness and content and how important that is and how TuSimple is focusing on that. This is a good time to dive into some of our simulation technologies as well as how we leverage them in our day-to-day autonomy and testing processes. First, I will give you a tour of our tools.
These tools help us turn real driving data into simulations as well as capture product and safety requirements and turn them into simulations through manual scenario creation processes. Next, I will introduce how we leverage our auto-triage pipeline to scale our validation and verification capabilities so that manual processes do not become a bottleneck when we start running at large scale these simulations every day. Okay, let's start with one of our primary simulation tools, regeneration. This tool utilizes the on-road recorded sensor data from our trucks. Once this data is recorded, our infrastructure allows the automated use of this data with newer versions of the software, which is especially useful for perception. Consider this particular case.
If you look very carefully in the LiDAR view on the left and compare it to the image on the right. I am hoping that the video will play. There we go. You will notice a particular false positive object when the video pauses. This is a common false positive object that could happen due to reflections of the road signs. This case is easily identified during the replay of the captured data with the results from the version of the software that was on board at the time. Right? Now, to see if this version, the new version of the software has improved and in fact solved the problem or not, we use our regeneration infrastructure to automatically compare against the past.
Essentially, regeneration is a time machine that lets our ADS experience any previously captured data at any moment in the past again to try a new version of the software. This tool is primarily powerful for perception and some prediction related issues. Okay, now let's look at another case study. This is a typical merge scenario. In a moment, you'll be seeing a ramp on the right-hand side from which a large vehicle will be merging into our lane on the highway. What happens next is our safety driver makes a precautionary disengagement and makes a lane change even though the ADS did not think it needed a lane change at here or nor it attempted it. How can we answer the question of what would have happened if the driver didn't take over?
The previous tool that I showed you is simply not enough on its own because it will not be able to create the correct relationship between the changing world state and the changing truck state to answer that question. This brings me to our next tool, which is ReSim, which targets an end-to-end simulation starting from perception results generated by the new version of the ADS on previously captured data, and it converts them into a reactive simulation to test the downstream. So here, what you're seeing is essentially the same video but actually generated the perception results generated with the newer version of the software. What we do is we take this and then we turn that into a planning simulation completely automatically. What you see here is very interesting. Notice that the case the driver did was a precautionary disengagement.
The ghost ego that you see on your screen is that particular motion. The solid ego is what the ADS would have done. As you can see, the ego, the solid ego, the ADS-driven ego is not actually making a lane change, but slowing its speed considerably so that it's able to let that vehicle merge in front of us. As you can see with this tool, we are able to bifurcate in time and see what would have happened in a parallel universe if the driver were not actually to disengage. Here's another typical case with pedestrians on the shoulder. This also results in a driver disengaging precautionarily to move away from the pedestrians and move on. Now let's watch the bifurcation in time to see what would have happened if the driver were not to disengage.
As you can see, the driver takes over the ghost ego, and at the expense of moving over the yellow line, they keep continuing. However, as our software do not break hard rules that are stated in the law, it stops for the pedestrians, and it would crawl behind them until there is enough space to safely bias away and keep going. So far, we talked about creating simulations from existing data. How about coverage, edge cases, requirements? Such scenarios are generally prescribed through specifications that are created by the edge cases in those specific requirements and then directly translated into planning simulations to test and answer the question of what would our ADS do in this particular situation. Here's a typical set of scenario properties resulting from requirements.
Sim Craft, our in-house tool, lets our scenario builders take these properties and turn them into a simulation in a matter of minutes. This is especially useful for creating edge cases and canonical driving situations in simulation and creating variations of them so that we can test all of the coverage that we know we will need in our ODD. Our final simulation tool that we will talk about is a rather interesting one, Meta Sim. The metaverse of simulations, if you will, where we can bifurcate time and space, not only for behavior that is affecting motion planning, but also for appearance, which impacts perception. All right. Lastly, I'd like to tell you a couple things about how we scale simulation. We generally where the bottleneck is proper evaluation of millions of simulations every day.
What is most important to note is in our ecosystem, we heavily rely on auto-triage tools that reduce the manual triage and result evaluation load considerably. As such, we can actually effectively deploy our simulation suite to play a primary role in our safety case analysis through virtual runs, which is essentially allowing for rapid virtual ADS experience to build validation confidence. Here are some typical numbers. We can run over 10 million mi in a day. Our infrastructure supports this, where our auto event detection recalls are all of the potential critical event candidates. Auto triage handles most of these events with high confidence. You're seeing a typical result from our validation for a driver outrun, where a fraction of them where it's unsure, the auto triage pipeline is unsure, is handled by manual triage.
Note that we can run this process efficiently anytime we update the software significantly, which reduces the need for large-scale road tests for every version of the software. With that, I'm going to pass it to Adrian, who will talk to you about systems and safety engineering.
Okay. Some of you have met me before. My name is Adrian Thompson. I'm the Vice President of Systems and Safety Engineering here at TuSimple. I've been at TuSimple for about a year. I lead the Product Systems Engineering, Safety Engineering, and System Verification and Validation teams. Prior to joining TuSimple, I've been the director of systems engineering at Waymo for driving behaviors, head of systems engineering and testing at Uber ATG. I've also been the director of systems engineering for multiple spacecraft, aircraft, and defense applications at L3Harris. In prior lives, I've also led military-embedded software and flight diagnostic development for GE Aviation and intelligent building systems and power systems development for United Technologies Research Center. That is a small sampling.
Overall, I have 30 years of experience developing complex software-intensive safety-critical systems across five major industries. I know we jumped forward, but so far you've heard from Xiaodi, Vince, and Ersin as they describe the wide array of safety features, functions, and innovations that we at TuSimple have implemented in our systems to achieve the high level of driver road safety we've achieved today. Now I'm going to spend some time introducing you to the overall safety perspective that has allowed us to achieve this level of safety rigor and that lays the groundwork for us to continue leading the industry in L4 trucking safety well into the future. The first thing we all need to agree on is that AV trucks are not robotaxis. Safety for Class 8 trucks is a completely different challenge.
These vehicles weigh up to 80,000 lbs when fully loaded. They take hundreds of meters to come to a complete stop, and they require more conservative steering and braking than smaller vehicles in order to prevent jackknifing or rear-end collisions. As such, our trucks have to anticipate situations far enough ahead to execute protective behaviors early, and must incorporate cascading layers of failure protections to ensure they retain the ability to respond safely at all times. We basically assert that the only way to achieve this level of safety is to design safety directly into the system from the ground up. We also need to keep in mind that it's not just about the autonomous truck. The entire AV operation must be safe.
For that reason, our safety approach is holistic and includes designing, verifying, and proving out every aspect of the autonomous operation, including the individual components like the base truck platform, the TuSimple autonomous driver, and also the Autonomous Freight Network as a whole. Additionally, our approach provides relevant guidance for the entire operational life cycle, from maintenance and pre-flight checks all the way through to how we manage and recover from failures. Now, unfortunately, there isn't a standard that adequately defines safety for autonomous trucking. Unlike the traditional automotive industry, which has mature standards and regulations, you know, that, you know, define exactly how to achieve safety, standards for the autonomous driving industry are in their early stages and are still evolving.
Consequently, in order to prove safety of our autonomous operations, we had to lead the way for the industry by adopting and improving best practices from traditional automotive standards, aerospace and defense standards, and several other standards from other safety-critical industries like medical, nuclear, et cetera. Now, combining all of these safety best practices resulted in what we call the TuSimple Safety Case Framework. The framework builds hierarchically, safety hierarchically by first ensuring that our systems are reliable and have a very, very low rate of failure. Ensuring that any failures that might occur, even at these very low rates, are handled safely. Ensuring that the driving abilities of the failure-free system are sufficient to complete the intended driving function safely. Finally, proving that the system is substantively achieving all of the above via extensive analysis, testing, and deployment in real-world conditions.
Using this framework, TuSimple has set the standard for safety in autonomous trucking. Now, one of the novel features of our safety case framework is our hierarchical validation of our autonomous trucks and our overall operations. Vince mentioned earlier how we do this with hardware, but what this really means across the board is that we always test every component, every subsystem, and then the system, and then finally the entire autonomous operation as we go along, making sure that we don't integrate any element into operation unless each one is individually proven to meet the applicable safety standard. TuSimple has also established an independent systems and safety engineering group with 60+ engineers and scientists possessing safety critical design and development experience. This group has been empowered to ensure compliance to our safety case framework.
One key role of this independent team is that it conducts adversarial testing to find weaknesses in the solution. You can think of this as similar to the way you might hire a renowned hacker to attempt to hack into your own information systems in order to identify any outstanding vulnerabilities. If any weaknesses are found in any of our adversarial tests, this independent team also holds the rest of the company accountable to fix them before approval to operate on public roads is granted. Now let's take a quick look at some examples of how our safety approach has been helping raise the safety bar in our system design.
One of the most important things an autonomous driving system does is control the steering of the vehicle to keep it in lane, to execute lane changes and safety shifts, like what we saw earlier, we call them biasing, or to even perform minimum risk maneuvers to safely stop the truck in the presence of failures. As part of our safety analysis, we identified all of the possible modes of failure that can occur within the steering system. We defined strict timing requirements for the system to fully switch from the primary steering channel to the secondary steering channel in order to remain safely in control at high speeds. As a result of this work, today, the redundant steering system on our trucks is able to switch from the primary to the backup channel in approximately 60 ms. Just like that.
Another really critical function of the autonomous driving system is keeping track of where the truck is on the high-definition map and in what direction it is heading. Precise position and orientation tracking is essential to making the right decision about what to do next, whether to stay in lane or avoid an imminent collision. Now we have always had a redundant approach available to measure our truck's position and orientation. During our test operations on the I-10 in Arizona, we came across a small number of locations where our confidence in our measurements was lower than desired. A root cause analysis of this problem using fault trees revealed that these locations all corresponded to places where both global positioning reception was compromised and few significant landmarks or surface features were available to anchor on visually or by LiDAR.
As a result, we mandated that a third method of position and orientation be implemented that doesn't suffer from the same weaknesses that the other methods were exhibiting in those locations. And as a result of that, we now have this third layer of localization and pose redundancy deployed in all of our autonomous vehicles. In conclusion, TuSimple has developed a rigorous and holistic approach to AV safety and is setting the standard for Class 8 Level 4 autonomy. We've also built a world-class systems and safety engineering group that independently and systematically verifies the safety of our overall autonomous trucking operation. Finally, we continue to update our driver safety case as the industry and our own designs and experience evolve to ensure that we are always fielding the safest and most capable autonomous trucks on public roads. That's the end of our technical presentation.
Thank you very much, and we'll be going to a five-minute break. Thanks.
Excuse me, if I could get your attention, if you could get back to your seat. Okay. All right. Welcome back, everybody. We've got a few more sections here, and we'll get towards the conclusion of today. The first one that we wanna go through is an update on our TRATON partnership. First, let's talk about TRATON. We work with TRATON's brands globally. Most notably, we work with the Scania brand in Europe, and Navistar in the United States. Our relationship with Navistar goes back several years. Predating even the TRATON/Navistar merger which happened last year.
Back in 2020, we did sign a joint development agreement with Navistar, and we think this is the first, the world's first independent collaboration between an AV technology company and a global truck OEM with the express purpose of developing a purpose-built L4 production truck. The mission of the partnership is to bring this Level 4 AV capability to the freight ecosystem at scale, reminding us about the third element of our development timeline. This is a significant undertaking. It's one where we partner with Navistar. We bring complementary skills, technologies, and experiences to the table. We're really thrilled to be continuing to deepen this partnership with Navistar, and we're marching forward on the path to develop the world's first L4 production semi truck. Next, we're gonna turn to a brief video that explores our partnership further.
The logistics industry is facing a number of challenges. The demands for freight movement are only increasing, as is the shortage of drivers to operate vehicles on the U.S. highway system. We can address this need by providing autonomous vehicles that don't require a driver and can operate with extreme safety and efficiency day and night, 24/7. TuSimple is a technology company focused specifically on autonomous trucking. We chose Navistar after exploring the entire field of possible OEM partners. Navistar is a leader in the development of Class 8 trucks, and they embrace innovation.
TuSimple brings in autonomy stack. Navistar brings in its hundreds of years of expertise in building trucks. This fusion is key to make this product successful.
For the past 20 or so years, Navistar has been focused on the commercial vehicle market and our vision for the future.
Navistar was willing to embrace the complexity involved in building a vehicle system from the ground up.
Building a truck from the ground up allows us to specifically address safe operation, efficient operation, and quality of product.
It allows us to put the right kinds of technology into the base vehicle system for the customers.
That's affirmative. We're in autonomy now.
The virtual driver is never distracted, never goes to sleep, doesn't need a break, can keep running constantly. Imagine a scenario where the driver drives the truck, drops it off at the terminal or the AFN network, and then the truck does the rest of the trip by itself till we picked up another driver who's waiting on the other end.
It allows us to provide safe, efficient, and sustainable autonomous freight movement for our customers, and it assures that you have faster and cheaper delivery of your products.
It's very important for Navistar to get to production with the first mass-produced autonomous vehicle.
We are seeing great success so far, both with our partnership and with the advancement of the technology.
It's a once in a lifetime opportunity to define where autonomous trucking can go.
I want to walk through some of the updates that we first talked about last week on our earnings call. The centerpiece of our Navistar partnership is the L4 production truck program. As I mentioned a slide or two ago, we announced this program back in mid-2020. What have we been doing for the last couple of years? Our teams on both sides have been working hand in hand to proceed through a gated automotive production process. This is a combination of technical and business work streams to develop a production truck that needs to be scalable, that can be placed into fleets around the country to provide safer, more environmentally friendly, and more efficient freight capacity.
One of the major work streams is working with the global supply base, and that's to both determine the overall vehicle architecture as well as to specify components for that production vehicle. This requires a lot of intricate planning, and it requires a lot of discussion and negotiation with the supply base. As we've been noting for the last several quarters, we have identified risks in the supply base, and we've been working to mitigate them. That's both to keep our product requirements to make sure that we meet those and to keep our timelines as tight as possible. One of the major risks that we identified starting about a year ago was the Autonomous Domain Controller.
As we discussed previously, we found a mitigation to this, which is also an opportunity for us, and that's that TuSimple is gonna be working with NVIDIA to design the ADC, and we'll work with a third-party manufacturer for production of that component. For other components, we continue to work with Navistar to both finalize the supplier timelines and incorporate them into our overall production plan. We're also very proud to announce earlier this month that Navistar's Escobedo, Mexico, production facility is gonna be the manufacturing site of our L4 production semi-truck. Today, Escobedo is Navistar's primary Class 8 production facility. It's a natural fit that our L4 vehicle is gonna be produced alongside all of Navistar's other Class 8 offerings. Finalizing this production facility selection is a major step, and we look forward to sharing more with you as we make preparations for that production process.
We also announced earlier this month that we expect to produce production intent prototype vehicles by 2024. We expect that these prototypes will be very close to our production vehicle. It's an important part of the pre-production process, and the prototypes will also be manufactured at the Escobedo plant. Just to take a little bit of a step back and talk about why this is a big step, we're gonna be going from retrofitting trucks in our Tucson facility, which is what we do today, to having a factory-built prototype.
That factory-built prototype is gonna be more reliable, more scalable, and we expect that we're gonna be able to start using these prototypes for traditional pre-production testing work but also for revenue-generating runs under our TuSimple capacity model. There's also a tremendous learnings by having a fleet of production intent prototypes, and that'll be substantial and it'll help our business model start to scale even before the start of serial production of sellable vehicles. In 2025, we are targeting fully integrated vehicles on the road. The exact start date of serial production of sellable vehicles is something that we are working on with Navistar, and we expect to share more precise details on that over the coming months. Serial production of sellable vehicles is really important.
That means that we will have a certified warrantied vehicle that is gonna be capable of being sold to third-party fleets, including the 15 blue-chip fleets that so far have placed reservations for approximately 7,500 production trucks. For these production vehicles, Navistar is gonna be responsible for the manufacturing and sale of those vehicles. But in order for those trucks to operate autonomously, we're gonna be charging a per mile virtual driver fee to enable autonomous operations on our HD maps. This is what we call our carrier-owned capacity business model that we talked about earlier. We expect that model to ramp quickly with the first substantial year of revenue contribution from this model expected to come in 2026.
As we take all these timeline changes into consideration, even with some of the delays, we do expect that we will be the first purpose-built production truck to the market. Our partnership with Navistar has never been stronger, and we are really excited about the future together. If we think about the production truck, this is really the tool that unlocks scale in our business model. It's not simple to produce a first of its kind sellable L4 vehicle that is capable of being produced at a scale of thousands and tens of thousands of trucks per year. We are very focused on achieving this goal together with our partners at Navistar, and we are committed with transparency with the market on our progress to do so.
To our knowledge, no AV trucking company so far has been as specific about their plans for a production vehicle that can be deployed at scale as TuSimple and Navistar. Building the first of anything is a complex process, and particularly for a technology as advanced as L4 AV driving. Despite that complexity, we're never gonna lose sight of the enormous opportunity in front of us. We believe that deploying L4 trucking at scale is going to be the most significant evolution of the freight ecosystem in decades. We are really pleased with the tremendous progress that we and Navistar have made together as partners, and we are very excited about the future ahead. Switching to the Europe region, we have a partnership with TRATON that has similarly deep roots to our Navistar partnership.
Also back in 2020, we announced a partnership with TRATON independent of our partnership with Navistar. This partnership had a focus on L4 trucking, both for the European and for the global markets. For those of you who don't know TRATON, it's one of the leading global truck OEMs, particularly in its home market of Europe, and it has leading brands including the Scania and MAN brands. We believe we could not ask for a stronger partner in this region than TRATON. Last year, we and TRATON together started the continent's first Level 4 terminal-to-terminal freight route in Europe. We began operating autonomous Scania trucks on Route E4, and that's a major freight corridor in Sweden. It runs from the north of the country down through Stockholm and into the southern region where truck freight gets connected to continental Europe.
In terms of next steps on the Europe partnership, we're gonna keep expanding the routes that we run on in addition to the E4 route in Sweden. Similar to our relationship with Navistar, we are intending to have a purpose-built truck under the Scania brand, and we expect to have more to announce about that production truck over the coming quarters. We're also gonna continue to judiciously build out our team in Europe and expand our presence in the region to support our partnership with TRATON. Europe's not our home market, but we still see tremendous opportunity for L4 trucking in the region, and we are pleased to have such a strong partner in TRATON GROUP to help us.
Our partnership with TRATON provides us the ability to leverage their existing brands, their presence, and their expertise in the region in order to bring these AV solutions to market. Our ultimate plan for Europe is intended to leverage these complementary strengths of us and TRATON to build the best solution and importantly, get to market quickly. We intend to take a capital-light approach in Europe. This is gonna leverage TRATON's long-standing presence and its network in the region. We expect to have more to announce in the coming quarters about this, but we generally expect that our partnership with TRATON in Europe will take the form of a technology licensing agreement where we provide support for the core AV functions while TRATON takes the lead on establishing an autonomous freight transportation solution for the customers. We think this is a win-win proposition.
It allows us to get to market fastest, and it plays to the strengths of each partner. The exact economic arrangements of such a partnership has not been announced yet, but we think we're gonna be able to strike an appropriate balance of taking this capital-light approach and develop an appropriate risk-reward profile for TuSimple. Overall, we're grateful for the partnership with TRATON. We believe the combination of TRATON's history of technology leadership, innovation, their presence in the region, combined with our world-leading L4 technology, is what it takes to realize our vision in the region. With that, we're gonna shift gears to the next step of our presentation, and I'm gonna invite Xiaodi to come up on stage with me. This part of the presentation is really about the impact of our driver-out operations and the path to commercialization.
We're gonna hand the mic back and forth. I'm gonna ask Xiaodi to take a seat on the stool, and then we'll go back and forth a little bit. Really this section is about merging both commercial and technical elements, and that's really the theme of the next couple of years here. We cannot have technology and commercialization in a vacuum. They have to be tied tightly together to use driver-out as a launchpad towards initial commercialization of our technology. Let's dive in a little bit and talk about that. Yeah, we're gonna do a Q&A at the end, Ken. Thanks. All right, we have a development framework at TuSimple of safety first, and then layering on efficiency, and finally adding scale.
Safety, as you've heard from Xiaodi's presentation, it means being able to take the driver out with our autonomous driving system, able to handle all of the situations safely. Our initial driver out operations are not efficient enough right now to be competitive in the commercial marketplace, so that's where the next element comes in, which is all about operational efficiency and driving towards commercialization. For efficiency, that means we're gonna remove certain elements from our driver out operations, like our support vehicles. We'll also improve our technology to improve productivity and expand the ODD. It's a fruitless exercise for us to develop a driver out vehicle if we cannot make it efficient enough to be competitive in the commercial marketplace, and we'll talk in a few minutes about how we measure that competitiveness. The last element that we've been talking about is scale.
Importantly, we think that we can achieve efficiency before we get to scale. This is a very important distinction. We think there is a significant advantage to achieving efficiency early. It creates a lot of discipline in the way that we develop our technology. For instance, we know that we cannot rely on inefficient operational elements, and one example of that would be a significant use of teleoperations. There may be a case where using extensive teleoperations for our driver out operations would be helpful to us, but we know that extensive teleoperations will never be efficient enough to be commercially viable. That's why our driver out program has had no teleoperation of any kind. Driving efficiency early is a critical element of our roadmap.
Another benefit of driving in efficiency early with our retrofitted trucks is that we can directly translate those learnings into our production truck program, ultimately creating a stronger business for the long term. What does commercialization mean? We think that in general, there are three elements to consider for any product to have commercial viability. First, at the top of the page you'll see you must charge a market price. I think in the freight market, it's actually pretty straightforward. Today, we operate our TuSimple capacity where our trucks are hauling freight for customers from point A to point B, and we're charging market rates. We do all of our commercial operations today using a safety driver, but we will start to expand our driver out program to do driver out commercial runs later this year.
Whether it's traditional manual freight, a driver in AV commercial run, or a driver out commercial run, the rate structure of those runs is always gonna reflect the market realities for freight on that route. Freight transportation is a B2B product, and it has a well-established and fairly visible pricing structure. If you move down to the second element here, we must meet the customer's expectations for the level of service. In truck freight, that means that we have to be able to operate on real freight lanes, and we have to meet the requisite level of service. That can mean pickup and delivery on time. You must have a certain availability of freight capacity for your customers. In AV trucking, that means that we need to be able to expand our ODD.
We need to be able to increase the reliability of our driver out operations so that we can keep a schedule. Third, you must have unit economics that make sense. This is universal, whether we're talking about a software business, a service business, or really any operation including AV trucking. You need a pathway to have profitable unit economics to show that your business makes sense. Trucking is a massive market. It has many hundreds of billions of miles run each year, but if we have no pathway to earn profits on a per mile basis, market size is irrelevant. That's why we are so focused today on the unit economics and specifically about our operating cost per mile. We have some headwinds today on operational cost efficiency. Today, we use retrofitted trucks. Those trucks have prototype components, and our operations are just getting started.
We must have a relentless focus starting now on driving our operational costs down per mile to a level where we are starting to be competitive with manually operated trucks. This is an important proof point, and it forces discipline into our development. You've heard me talk about our production truck program, our path to commercialization, and doing that path to commercialization even ahead of a production truck. On this slide, what we're trying to do is pull it all together with one driver out evolution timeline. First, to set the stage, at the end of 2021, we started driver out testing operations, and those testing operations for a driver out vehicle are gonna continue indefinitely. That's because we will always need a testing environment to develop and refine our technology.
Today, we run our driver out tests on semi-trucks that we retrofit here in our facility in Tucson. We don't yet haul commercial freight driver out, but as we've mentioned before, we will start to later this year with Union Pacific. Today, our driver out fleet is relatively small. It numbers approximately 10 vehicles. As we've talked about, we run fairly limited operations, so far only at night on a single route in Arizona. As we progress through the rest of this year and then advance into 2023, we are gonna expand our driver out operations. That'll be more trucks running day and night in Arizona and Texas. Our operations are gonna get more efficient, particularly as we remove the chase vehicles.
We will still be running on retrofitted trucks this year and next year, but we expect by the end of 2023, even with those retrofitted trucks, we're gonna have an operating cost per mile that is starting to be competitive with human operations. At that point, we expect that our driver out commercial operations is gonna be in full swing on TuSimple capacity business model. Which again, just as a reminder, means that we own or lease the trucks, but we are hauling freight driver out for our customers, earning revenue. At this point in the overall evolution, we think we will have reached initial commercialization. As a reminder, that means we'll be charging a market rate, we'll be running real freight routes, keeping a real freight schedule, and meeting the required level of service of our customers.
We'll also have unit economics that are starting to make sense and have a clear line of sight to be competitive in the marketplace at this point. Importantly, just to make sure there's no confusion, this commercialization is an inflection point. It's not the destination. We are planning to scale our driver out operations further. In 2024, that means that we're gonna start having the opportunity to begin using production intent prototypes built in the Navistar facility in Escobedo, Mexico. That is a big point of distinction, instead of using retrofitted trucks that are built by us and retrofitted in our Tucson facility. At that point, we won't have a certified warrantied vehicle for sale to customers, but we will be able to drive further efficiency, and we'll be able to scale into the hundreds of trucks on the TuSimple capacity model.
As we start to move into 2025 and 2026, and we have a sellable production vehicle, we unlock the carrier-owned capacity business model, and that's where our fleet customers can own the vehicles and pay us the virtual driver fee for driver out operations. At this stage, with both models unlocked, we can significantly scale the operation into thousands and ultimately tens of thousands of trucks that are coming onto our network each year. The production truck's reliability and the economies of scale will continue to drive further operating cost reductions to the point that AV trucks will have a significant cost advantage, and we can share that cost advantage with our customers and the rest of the freight ecosystem. All the while, we're gonna build a highly attractive, large and defensible business.
We're gonna dive into the details, but this slide is important in order to set the stage on the evolution on where we are today and where we expect to go to by the middle part of this decade. In terms of where we're gonna expand our driver out operations, we have deliberately chosen our first set of driver out routes to be in OD pairs in key freight locations, both in Arizona but also expanding into the Texas triangle. Based on the discussions that we have with our partners and our customers, and taking into account some of the ODD and regulatory considerations, we believe that these two areas are the most conducive for initial driver out commercial runs. As we mentioned earlier today, there is a significant market size associated with these initial routes. These are real market opportunities for us immediately.
As announced, we're gonna start hauling driver out freight with Union Pacific starting this year, and we are expanding our ODD to include rail intermodal containers, which opens up more market opportunity for us. We intend to build upon that initial driver out customer operation with other similar runs with other customers as we start to move through 2022 and 2023. Then we'll systematically and strategically add more driver out runs based on ODD, customer demand, lane density as time progresses, and broaden out the driver out reach on our AFN. The goal of our initial driver out commercial operation is to drive efficiency, and we need to do that and to start preparing for scale. A testing environment is always gonna be critical for the development of our technology. Ultimately, we are focused on putting our technology into commercial use cases.
That helps to spur adoption. It helps to share the benefits across the ecosystem. We've had a lot of talks about efficiency. How do we actually measure efficiency and see how we're making progress to commercial viability? No one else in the industry has taken the driver out yet. This is a very novel question that we had to ask ourselves. Some basics about how we are approaching this from a conceptual perspective. We need to measure against the baseline. We start with the baseline of a current truck freight solution, which is the human-operated truck that operates today. There are lots of estimates about operating costs, whether it's from ATRI or the National Private Truck Council. We took a look at these industry figures. We tried to reflect the current fuel cost environment, although that's changing very quickly under our feet, as we know.
We tried to estimate a baseline manually driven truck, and we estimate that to be roughly $2.50 per mile. Importantly, we know this baseline cost is gonna move around. It's gonna be higher or lower over time. It'll depend upon the use case, the region and the overall market environment, and we'll continue to monitor that and adjust it as needed. We take that baseline, and we compare it against what we assess to be our driver out operational costs. What are our driver out costs? There are some costs that are gonna be the same as the base cost. Think about something like tolls and permits. Then there are a lot of costs that are gonna be incremental. Think about extra maintenance and calibration costs or added vehicle hardware expenses for that driver out vehicle.
Importantly, we are including the operational costs, but we are excluding the true development costs. I will acknowledge this requires some real judgment on what is an operational cost and what is a development cost. As we ramp up our driver out, where we get more runs and more miles of data, we'll continue to refine this methodology. We are committed to getting the most accurate measurement of our driver out operational costs, both internally and what we report externally. This is gonna be a very important operating metric for us, and we actually hope for the rest of the industry as well. We will continue to report that and hold ourselves accountable to measure progress. Let's dive into the categories of operational cost. First is the driver cost, whether you're talking about a virtual driver or a human driver.
Let's think about what actually comprises a virtual driver cost. Virtual driver has items like data storage, transmission, remote monitoring, the maintenance of our HD map network. As a reminder, we're not including the development cost for that virtual driver, just the fully loaded ongoing operational cost. For AV operations. In addition to our base vehicle cost, our driver out vehicles are gonna have very AV specific hardware. That's the things that you've seen in taking your test ride. It's the sensor suite, the compute units, the cabling, all of the extra electrical components. For all of these vehicle costs, we need to compare the ongoing operational charges such as the cost of that AV hardware spread over its operating life. Next, there's maintenance costs.
In addition to the traditional maintenance of the base vehicle that every fleet has to deal with today, we have included incremental AV equipment-specific maintenance. That can include things like the cost to periodically recalibrate the sensor suite. Due to natural shifts in the sensor equipment, we do have to do this periodically. It also includes the cost to rescue the AV vehicle. You've heard us talk about minimum risk condition, and Xiaodi will talk more about it in a bit. If there's an event where there is a system degradation, our truck has to pull over to the side of the road, and that mission cannot be remotely restarted, we'll have to send a rescue driver out in order to complete the mission. That is an operational cost, and we intend to incorporate it into our estimates of our operating cost per mile.
Next, for fuel. Today, we have the exact same propulsion system as the vast majority of the human-operated trucks that are on the road. As you've heard us talk about, we do expect to have an improved fuel economy that we've been able to validate both on a university study and in our work with UPS at approximately a 10% advantage based on those studies. Lastly, there are other operational costs. These are things like tolls, permits, and vehicle insurance. The first two of those are identical between AV and human operations, but for insurance, we are modeling in an increase in insurance costs for our initial driver out runs.
We think this is appropriate and conservative, and these rates that we are modeling in here are informed by the rates that we are actually achieving in the marketplace for the initial runs that we started in December of 2021. Now I'm gonna hand it over to Xiaodi, and he's gonna talk about how some of the technical progress that he and the technical team are making will immediately and over time start to drive some of these efficiencies that I was talking about.
Thank you, Pat. Okay. We wanted to level set our progress relative to the market. TuSimple is the only company operating driverless heavy duty trucks, and is the only autonomous driving company in the U.S. operating at a highway speed. One important note is that we do not rely on any teleoperation in our driver outs. Reliance on teleoperation is a difficult cost to optimize. Now, we believe that our approach really cleared a lot of doubts on the road to commercialize. Also, it is important to note that in trucking, getting to driver out is a more significant milestone comparing to that of robotaxi. As we think about our portion of the AV trucking problem that is related to the ODD, trucking is about maneuvering a more complex vehicle in a more defined problem relative to passenger car applications.
In terms of measuring from driver out as a waypoint to actual commercialization, we think trucking is relatively nearer term comparing to that of robotaxi. As Pat has mentioned, there will be some important evolutions this year and next year in our driver out operations. First, we will remove the support vehicle, including both the chase van and the survey vehicle. By the end of next year, we expect to run driver out operation without any support vehicle. Second, we will add day operation to our current night operations for driver outs. This is critical to expand our operation to closer to 24 hours and make sure that our driver out operation meets the required level of service from our customers.
Third, all the while, we will expand availability of our driver out operations as we make our system more resilient and make our maintenance and calibration more efficient. Fourth, we will expand driver out operation into Texas so that we have multiple real freight routes and running on driver out basis. Finally, while today, we operate driver out only in the testing mode, by the end of next year, we will run customer freights in driver out mode. It is probably helpful to spend some time here to review the driver out program that we have done last year. First of all, in our driver out program, we have a survey vehicle. This is to make sure that there is no extreme scenarios outside our current operational design domain, such as complete road closure with police interaction.
The survey vehicle stay 5 mi-6 mi ahead of our autonomous truck. Second, we use a chase van that follows the autonomous truck. There is a small probability of hardware failures during the mission. Under that circumstance, our autonomous truck will abort the mission, stop safely, and the backup driver and the backup test engineer in the van will rescue the autonomous truck. The chase van always stays behind the autonomous truck to guarantee zero interference with the traffic near the autonomous truck. Third, the unmarked law enforcement is there. Our driver out operation is closely monitored by the Arizona Department of Public Safety. To minimize any possible interference, we requested the law enforcement vehicle to be unmarked. You know that you always drive more politely near a police car, so that we wanted to minimize that impact as well.
Furthermore, the police vehicles are kept at least half a mile behind the truck to reduce the impact on the traffic. Starting in this year and before the end of next year, we will replace the lead vehicle by another truck in the fleet, and adoption of collaborative mapping is really helpful for that. In the end, one of the advantages of operating autonomous freight fleets is that there's a lot of collaboration between different trucks on the same route. With the capability of aggregating on-road information to mapping in almost real-time, the entire fleet will be notified if there's anything unusual on the roadway. We also plan to remove the chase van by completing three tasks. One, upgrading our truck with more reliable hardware so that we have less probability to trigger the MRC. Second, to establish a roadside assistance program to us.
For example, the Werner Enterprises Roadside Assistance is a choice. Three, t o build a teleoperations oversight command center that can monitor the health of all trucks from the cloud to better coordinate the roadside assistance if needed. Once we completed all of these steps, we will be much more ready to the fleet level commercial operation of our driver out trucks. Based on the initial success of driver out, in this quarter, we focus on enhancing and polishing the system for the minimal risk condition, and I'm proud to report our current progress on this critical capability. We all know that hardware failure is inevitable, but what our truck can do now is to keep driving with a partial system up to 10 minutes until it finds an appropriate spot on the roadside. This is a critical prerequisite capability for us to remove the chase van in the future.
Let's take a look at our video. This is a video footage of our truck on the public highway in Tucson. As designed, our truck receives a mock-up signal from a system failure, therefore entering into MRC state and keeps at the rightmost lane, and then find available areas, that is empty and then slow down to a stop at the roadside, like we see here. Now I'll hand it back to Pat to go through some of the detailed numbers. Thank you, Pat.
Great. Thanks, Xiaodi. Now we're gonna pull this together and take a look at the progression of our unit economics. You heard me say earlier that the current cost is not competitive, and we peg it at about $35-$40 per mile today, which is really high. You can see though, in this chart that the vast majority of our driver out operating costs today is driven by our support vehicles, and which we've said we intend to remove from our operations by the end of next year. Further, as we go over the course of 2022 and 2023, we intend to have improvements in utilization. That means that we're gonna have more driver out operating hours per day, plus you couple that with the natural maturation process of key hardware and remote monitoring capabilities.
These are expected to bring down our cost per mile to mid-single digits by the end of the next year. This is a significant reduction in that metric in a relatively short period of time. When you think about it's actually quite intuitive that if you're able to remove the support vehicles and you're able to add some basic efficiency, it can be quite impactful to bringing down the cost per mile. As Xiaodi talked about and showed on the video, removing the support vehicle requires some advanced capabilities like our MRC capability, and we are very confident in our ability to remove the support vehicles over the next 18 or so months. As we are gonna progress with our production truck program in 2024 and 2025, we're gonna have a significant reliability benefit that comes with having automotive grade or near automotive grade components.
You start to combine that reliability with the economies of scale from a global procurement process, and that's gonna help to drive down the cost rapidly even further. At this stage, we also expect to see further utilization improvement. At this point, we believe that our AV operating costs will be significantly reduced from that of a manually operated truck. That's what we think about as the run rate end state, where we're gonna be able to unlock meaningful savings for our customers, and we're gonna be able to develop a very attractive margin profile for our business. Over the next couple of slides, we'll go through some of the details about the specific cost elements. First, let's talk about support vehicles. This one is fairly straightforward in the fact that we will remove it over the next 18 months.
Just to think about why is that figure so high? The costs that are included in this are the vehicles, the advanced equipment that we put on them, the fuel that we use to operate them, and the labor of the crews that are inside each of these support vehicles. Just for context, the cost of our support vehicles for a single day is about $4,000-$5,000 per day. Today, we run two runs per night or only about 160 mi. That translates into a very expensive cost per mile that is not commercially viable. Once we're able to take these vehicles out, it has a significant reduction in the cost per mile.
This will be the single biggest reduction to our operating cost per mile from where we are today to where we expect to be by the end of 2023. Next, let's talk about the virtual driver costs. You've heard Xiaodi talk before that it's not just removing the human driver and that there's no operating expense to replace that driver. There are a series of costs that you need to take into account, and you need to drive efficiency on them. Specific costs for the virtual driver. First one is mapping maintenance. We have a network of HD map routes, and those require maintenance in order to be able to operate on those map routes.
A lot of the work that Robert Rossi and his team are doing every day is making our mapping more efficient, reducing cost, as well as reducing latency and reducing the amount of time between updates. This is an area of critical focus for us. These costs also include data transmission and storage costs. There's a ton of data that we collect on our trucks, but we have to prioritize which data we need to send back in order to help us with our AV operations in real time, as well as to manage the overall costs from the data transmission and storage costs. Lastly, our virtual driver costs also include our remote monitors. Think about these as sort of pseudo air traffic controllers. They need to monitor where each one of our trucks are in our network.
If there were to be something that's beyond the horizon, so think about, like, an extreme weather event, they can actually put the truck into MRC to wait for that extreme weather event to clear. Over time, we intend to have the ratio of AV semi-trucks to remote monitors increase, which will drive operational improvement in terms of our operating cost per mile. When you think about the vehicle cost, an AV truck has significant incremental costs that we need to take into account. That's the sophisticated sensors, the compute unit, enhanced wiring, electrical systems, redundant braking and steering, and more. As we get into more advanced generations of retrofitted trucks, and then eventually to prototypes and into the production vehicle, we're gonna be able to bring down the cost per mile for these elements.
This is gonna be because we'll have automotive-grade components that will increase our reliability and our uptime, and we'll also have an OEM procurement process that'll reduce the cost per mile of these. Thinking about the specific elements, there are a couple big ones that we would highlight. The first one is that we have been working on the Autonomous Domain Controller, and you heard us talk a lot about that. What does that mean practically? Well, it is actually significantly less expensive than the servers that we use in the trucks that you've all been riding in in your visit to Tucson, and we think it'll be a significantly more efficient, and more reliable than those servers that we use today.
We also expect that as we get to the natural maturation of key components by the supply base, going from A sample to B sample to C sample to ultimately automotive-grade components, that will also drive incremental cost improvement. Key components are LiDAR and the remainder of the sensor suite, braking systems, steering systems, electrical systems. As we talked about, as we get to the economies of scale and we harness the power of the global procurement, departments of TRATON GROUP, we're gonna be able to drive further cost reductions as we get to that production vehicle. Last one, we're gonna talk about some of our other operating costs. First one here is thinking about fuel costs. We are incorporating our fuel efficiency, which we have demonstrated in our university study and our work with UPS.
For insurance, as I talked about earlier, we are assuming a higher insurance cost for driver out, you know, and that is based upon the rates that we're seeing in the market in our discussion with insurance carriers today. We are optimistic that over time, we will be able to drive down the insurance cost per mile as we have more data, and we're able to prove out the safety for our system from an insurance perspective. This also includes the cost for maintenance and calibration. These events will reduce considerably as we incorporate more reliable software and hardware, and we couple it with continuous improvements in our technology overall. As we talked about earlier, we need to take into account the operational cost of maintaining that AV system if we are really gonna measure the progress towards commercialization.
System reliability issues have a cost, and we need to incorporate those into our model. For all of our operating costs, it's also about increasing the utilization of our system and running more miles per day. By running more miles per day, we're able to increase the denominator in this equation. Sometimes it's easy to be overly focused on the numerator, but the denominator is a critical part of driving the operational efficiency. Let's wrap this up a little bit. In summary, we have an operating cost of $35-$40 per mile today. It's uncompetitive with the marketplace. These operational milestones that we laid out starting at the beginning of this year that we expect to all achieve by the end of next year will rapidly bring down that cost to mid-single digits.
Removing the support vehicles, increasing utilization, and the other improvements that I talked about. At these levels, we are gonna start to approach competitiveness with the commercial market on an operating cost per mile, and as we progress to the scale operations with prototypes and eventually the production truck, we're gonna bring that cost level down to a level well below a human-operated truck. Once you have this cost advantage and you're deploying it at the scale of thousands and tens of thousands of trucks, it's gonna unlock the full economic potential of our AFN, both for our customers, but also for TuSimple. How are we gonna track progress? We pride ourselves on trying to be the most transparent AV company. With all the information we've provided today, we wanna make sure that you can hold us accountable for actually making progress on these commercialization goals.
We're gonna intend to provide you regular updates on the evolution of our driver out operations. Every quarter, we're gonna provide any applicable updates on our driver out routes and our driver out fleet size. As we have updates that are shareable on customer participation in driver out programs, we're gonna share those as well on an ad hoc basis. The evolution of the hardware that we're using in the driver out program will likely be very closely linked to some of the progress we're making on our production truck program, and we're gonna intend to provide updates on an ad hoc basis whenever it's appropriate. The last one at the bottom, but an important one, is that we will continue to provide updates on our progress in reducing our operational cost per mile.
We expect to provide an annual update at the end of 2022 and the end of 2023, and then as we start to have more data about our driver out program, we'll provide more regular updates thereafter. We're gonna wrap this up by translating this into a financial update, including some of our long-term financial targets. First, before we get into the financial targets, I thought I'd level set a little bit about the total addressable market as well as the go-to-market strategy by each region. First, in our home market of the U.S., it's very well known this is a large market, approximately $800 billion in terms of total market size for the truck freight market. Our go-to-market strategy is dual-pronged.
We're gonna be running both the TuSimple capacity model, where we own or lease our own trucks and haul customer freight, as well as the carrier-owned capacity model. The carrier-owned capacity model is where our customers can buy the truck from Navistar and pay us for our white glove virtual driver service to be able to place those trucks into their own network. In Europe, this is a $400 billion market in its own right. As we talked about earlier, we're gonna partner with TRATON in a capital-light approach for this. We think this makes sense because it allows us to get to Europe quickly without overstretching both our capital capacity as well as our bandwidth from our management team. It allows us to leverage the strengths that TRATON has in the market.
In China, as I talked about at an earlier session or an earlier section, it is the world's second-biggest freight market, so it's a massive opportunity. We do have multiple paths to market, both with an L2 solution and eventually an L4 solution with an AFN that is very similar to what we are doing in the United States. I would note that with China, we also see an opportunity to expand our technology and our commercialization to other regions in Greater Asia as well. Let's review the unit economics. We talked a few minutes ago about the power of driving efficiency and scale to unlock the economic potential of the AFN. That benefits both our customers as well as us.
To think about the unit economics here, first on the top part of this page, let's think about the revenue profile to TuSimple from a scaled production truck. As you see on the top left, if you assume 150,000 mi-225,000 mi of operation per year and prevailing market freight rates of $2.25-$2.50, each one of our TuSimple capacity trucks can earn approximately $340,000-$560,000 of annual revenue. These are illustrative run rate figures. The one thing I would say around the utilization is that we are confident that as our technology matures, our semi-trucks are gonna be very high utilization assets. Our virtual driver is not subject to any hours of service regulations that a human driver is subject to today.
We expect these trucks to be in near continuous operation as they get put out into the network. For carrier-owned capacity, which is the second line here, we have not yet established any pricing with our customer base. Given the prevailing driver wages, which can be well in excess of $1 per mile, we've listed an illustrative $0.45-$0.65 rate for the virtual driver service, and that is revenue that comes to TuSimple. We believe this will create a very attractive top-line business for us on the carrier-owned capacity model. These, at these pricing levels, incorporating those same utilization expectations, that'll imply $67,000-$146,000 of annual revenue to TuSimple for each truck that we put out on the carrier-owned capacity model.
More importantly, though, the benefits to our customers is very significant under each model. For TuSimple capacity, where we're hauling freight for shippers from point A to point B, we expect to offer our customers a differentiated level of service. We're able to drive 20+ hours per day, which opens up the network for rapid freight moves. We'll have the enhanced safety, reliability, and meaningfully improved environmental footprint that comes with Level 4 AV trucking. The efficiency of our technology is gonna provide a lot of flexibility in our cost structure so that we can price competitively in the marketplace for TuSimple capacity. For the bottom row here, you see for carrier-owned capacity. We've put some illustrative estimates of the per mile savings of $0.35-$0.55 per mile.
The actual savings will depend on our pricing that we establish with our customers, and it'll also be relative to the human cost of a driver. We know that there are human costs, the human cost of a driver can be well in excess of $1 per mile, and we've seen several shippers that have noted that publicly recently. We expect that customer saving, the savings regardless of what the baseline human cost is to be very substantial. We've pegged it here using these utilization figures at $260,000-$620,000 potential net savings over the useful life of any single truck that goes into our customer's network.
That does not take into account the fact that the purchase price for the Navistar L4 production truck will be higher than a typical manually driven truck, but we do expect that the payback period on that incremental CapEx will be measured in months. It'll be a very quick payback period. If you take these savings, even net of the incremental CapEx, and you multiply that across thousands or even more, for large fleets, it's a really significant economic opportunity for our customers in addition to all of the safety, reliability, and environmental benefits that I talked about earlier. We operate in a market that is heavily influenced by the spot rate for freight transportation. We know the prevailing driver wages can have a degree of volatility, and we know that the spot rate will move up and down. These figures are illustrative.
They are gonna move in a variety of ways over the coming years. The overall takeaways, no matter what, remain the same. Our technology and our business model is gonna allow us to provide highly compelling value and unit economics to our customers, and it's gonna build a really attractive and defensible business model for TuSimple. Those unit economics are hyper-compelling. How do we see our business scaling as our production trucks start to get adopted? This page that you see here is focused on the number of trucks on the AFN. These are the average trucks that are on our AFN in each one of these years of 2026, 2028, and 2030. As a reminder, we don't benefit from the sale of a truck in the carrier-owned capacity model. The vast majority of the sale of that truck remains with Navistar.
How do we earn revenue? Well, we earn revenue by the autonomous miles that are operated by the trucks that are on our network. For us, the strategic goal is to get trucks onto the AFN and then increase the utilization of those autonomous trucks as driving our future revenue under both models. 2026, as we talked about earlier, we expect to be the first year of significant contribution from the carrier-owned capacity trucks. That's because that's when the Navistar production program will be in full swing and ramping up. Similarly, in that year of 2026, because of the scalability of the production truck, we expect to be able to take more trucks into the TuSimple capacity fleet and be able to ramp up the revenue associated with that model as well.
As you can see, most of the revenue on this top chart and most of the trucks are coming from our U.S. operations. But we're also gonna ramp up the truck fleet in our international operations from our China business, as well as from our European business that we intend to build with our partner TRATON. We see this growing rapidly from 2026 to 2028 to 2030 as the production program ramps up and the economic benefits and other benefits of our technology are realized in the freight ecosystem. We expect to end by the end of 2030 with over 50,000 trucks on our network globally.
In terms of revenue, in 2026, we expect over $500 million of revenue, and we expect that to grow rapidly to $3 billion-$4 billion of revenue in 2028 and over $10 billion by 2030 from our global operation. The ramp, as you would expect, it starts to ramp significantly in 2026 after the production truck program begins to scale up, and that's when we start to have the ability to put thousands and tens of thousands of trucks onto our network in any given year. Importantly, I would say we have a lot of flexibility and confidence in this 2026 revenue figure.
With an earlier start date of serial production of sellable trucks, we can get more trucks into the hands of our carrier-owned capacity customers, but we also retain the flexibility to be able to ramp up TuSimple capacity using prototype trucks as well. As we scale up, we continue to get more and more efficient, and we drive a more and more attractive margin profile as we move through time. That's on a blended basis across all of our geographies and the two primary business models in the United States. Throughout this time period, we expect to see the utilization of our trucks continue to increase. I would note that our forecasts are not including any kind of major increases in the rates per mile that we charge for TuSimple capacity or the rate that we charge for carrier-owned capacity over this time period.
With that, I'm gonna hand it to Xiaodi, where he's gonna give some closing remarks, and then we'll come back and do Q&A so we can address the questions.
Thank you, Pat. Finally, I wanted to turn to what all of this means in terms of our overall business strategy and strategic positioning. I would like to close our presentation by reemphasizing that the path to commercialization is a straight line that passes through driver-outs. If we wanted to commercialize, we must not only solve the autonomy software development problem, but also the complex validation and ruggedization problem that arises when taking the safety driver out of the truck for both hardware and software aspects. As we have shown you in our technology section, we are the first in the world to develop a driver-out framework for high-speed driving. The know-how and strategic advantage we have through this framework essentially put us into commercialization, the second episode of the trilogy.
We have been the first to start, first to develop and successfully implement a driver-out framework, and we are confident that we will be the first to commercialize by rapidly reducing the cost and migrating our operating domains. As we continue to set the pace for the industry, we fully expect to be the first to market, which has a significant advantage in an industry that will be profoundly transformed by such an impactful technology. Having a scalable and efficient approach means that we will have a strong and defensible long-term margin profile. Bringing this technology to market via our Autonomous Freight Network builds deep ties with our customers and other partners as we share the immense value proposition across ecosystems. This builds a highly attractive and defensible business model. We will have a five-minute break, and after that we'll start the Q&A. Thank you.
All right, we're gonna get started with Q&A. We have two mics that we're running through. If I could just ask for you to wait for the mic, say your name for the webcast, and then your question. I really appreciate it. All right. Ravi.
Thank you. Ravi Shanker, Morgan Stanley. Gentlemen, thank you very much for the detail. If I can start with a couple of questions for Pat on the numbers. The $2.50 a mile that you're assuming for human driven and the $1.50 you're assuming for your number. I'm assuming the $1.50 is not super tight to the cycle. Let's say in 2025, the $2.50 is like $4.
Yeah.
Is the $1.50 gonna be close to?
Yeah, I mean, I think it's. You can think of it as the relative positioning between the $1.50 and the $2.50. We think we have an ability to be in that, you know, that relatively a third cheaper or more relative to human operation. A lot depends on the composition of what the cost structure is, whether it be the fuel cost and what happens with fuel rates over time. The most important takeaway is that there will be a significant reduction by the time we get to scaled production of our AV truck.
Got it. You had said before that you don't anticipate needing to raise any more capital until the start of commercial production. Now that's pushed out a little bit longer. Kind of what do you think about that?
Yeah. I understand the nature of the question. I mean, I guess I would just emphasize we have $1.25 billion on our balance sheet. No significant liabilities. Our cash burn for the year, you know, we said we would expect to end with it about $900 million. We think there's actually a lot of levers that we can pull to both reduce the burn by being more efficient, and you're starting to see some of the fruits of that labor. We also have the potential Asia-Pacific transaction. It's too preliminary to know what that would entail, but we did note on our first quarter earnings call last week that approximately 20% of our EBITDA loss does come from that business.
If you think about a range of potential outcomes where we could separately fund the development costs for that business, as well as the potential to actually take proceeds back to TuSimple, it gives us some potential to extend that runway as well. It's just too preliminary for me to be able to handicap exactly what that would be. We do feel really good about where we stand. We're really grateful for the capital that we have on our balance sheet. We think there's a number of different levers that we can pull in order to extend our runway. We feel like we're in a, you know, a pretty privileged position in terms of the balance sheet strength that we have.
We know that's important in terms of how we interact with our partners, our employees, et cetera, and allows us to stay very long-term focused on completing the vision as opposed to having to take any kind of unnatural deviations on the technology development.
Got it. Just lastly, one question on Xiaodi. Maybe this is a big picture question or a technology question. One of your competitors is, or at least two of them are trying to do both passenger car and trucking at the same time with virtually the same tech stack and the same solution. What do you think about that? Again, we just saw that you have a lot of work to do for the next 10 years, so I know that none of you are gonna get any sleep. But at some point in the future, do you think that it is possible to use your current tech stack for the passenger car business over time? Thank you.
Okay. I think first of all, the biggest difference between passenger cars and trucks is like high-speed driving. In order to build a high-speed driving capable suite, you have to develop a lot of things like from scratch. I don't believe that many of these are easily transferable in near future. Maybe in far future, maybe 10 years later, when the technology is truly democratized, which meaning that a college student has their kind of course project due to autonomous driving vehicles. At those times, which is like, you know, everyone can build a cell phone easily, those are the times when things can democratize. When they come to those times, technology advantage is not going to be the advantage.
Great. Thanks. Chris Wetherbee from Citi. Thanks for the presentation, guys. It was very informative. Maybe a couple of questions to start for Pat. Just wanna get a sense, I think you mentioned $5-$6 per mile as a competitive dynamic. I just wanna make sure I understood how you were thinking about that $5-$6 relative to the competition.
Yeah. I would say it's approaching competitiveness, right? If we were running our TuSimple capacity permanently at $5-$6 per mile, we would be permanently the most unprofitable trucking company out there. I understand that. You know, the point that we've used, whether it's approaching competitiveness or line of sight, the point is that we're measuring it and we're trying to give, you know, our most accurate expectation about where we'll be at the end of 2023.
The elements that will help get us from that $5-$6 per mile down to a $1.50 per mile, towards the right-hand side of that chart, those are all things that are also in process where you can kind of clearly see how they will translate to further reductions, where you're starting to get to a cost that's at parity with the human and then below that eventually. That's the things around the hardware development. But, you know, what we wanted to avoid is just punting and saying, "Yeah, we're using retrofitted trucks right now. Don't worry about the cost structure. Once we have a production truck, that'll solve everything." That's not an efficient approach. First of all, you won't be able to engineer efficiency into the production truck if you don't start engineering efficiency into the retrofitted truck today.
It creates that discipline. It also helps us to develop that commercial muscle to be able to actually commercialize and haul real freight on a driver-out basis. I mean, that's what this is really about, is starting from one or two or three lanes, being able to actually have real commercial operations where we don't have all of these unnecessary, or I should say, non-scalable, non-economic, kind of aids or crutches that help us do that, but never will turn into a business model. For that remainder, going from $5-$6 down to $1.50, we have to have a very clear line of sight that that will happen over the course of the ensuing couple of years after that.
Okay. That's very helpful. Just on the retrofit side, I think on the timeline.
Mm-hmm.
Slide that you laid out, there was hundreds of retrofits. I just wanted to get a sense, when does that happen? Do you have the capabilities in-house to be able to build hundreds of retrofits today?
Well, I would say our global fleet today is about 100 trucks. There's 75 or so in the U.S. That'll increase. Will we have 100 retrofitted driver-out trucks? I think you could get to that level. Where you start to get into multiple hundreds is when you start to take the prototype truck that's built at the Navistar facility in Escobedo, because it's coming off of a factory line. You could have those prototype trucks, which are production intent, they're coming off the line. Some are certainly gonna be used in typical pre-production testing, but some could actually come into our TuSimple capacity operations. They won't be sellable to a third party, but we'll be able to self-operate those and put those into the AFN and be able to haul customer freight.
That's when you can get into hundreds of trucks on that TuSimple capacity. That's sort of the mesh point where you start to go from 100 or so to multiple hundreds. Then it's really when you get that serial production of a sellable vehicle that's coming off the line in Escobedo, that's when you can start to get to the thousands and tens of thousands.
Just the last question is on that. In terms of how many trucks you think you need to really get the economies of scale and that price down to about $1.50, is it that first year where you have between 2,000 or maybe 3,000 trucks on the road, or does it need more scale than that?
No, I think that's the general point where you would start to see that scale coming down, where if you've got 1,000 trucks that are running on TuSimple capacity, I mean, that's a fairly sizable operation from a fleet perspective. So there's always a lot of variables, but that would be our intent, is it's not waiting until you get out to 2030 to get to that level. It would be really unlocked pretty closely thereafter when you get to that initial serial production of a sellable vehicle.
Hey, it's Ken Hoexter from B of A. Just, Pat, just a clarification on one of your slides there. You had $2.50 cost per mile. You saw rates in the market at $2.25-$2.50. Just to understand, is the industry losing money today on that mix?
No. I mean, I think what we're trying to show is the rates that we're achieving in the Southwest, where we run, which is a cheaper freight region than where you would see nationwide. If you actually look, I mean, from the ATRI data that we would see, you know, maybe the average rate is $2.75 or $3 per mile versus an operating cost of $2.50. It's sort of a fool's errand to kind of say that there's a standard freight rate across the country, across different use cases, and similarly, that there's a standard operating cost per mile. It really just depends. You know, I knew somebody would ask me about this question, so it's a good one to ask about it.
You know, the reality is, what we didn't wanna do is say, just because the nationwide rate is $3 per mile, and we all see trends that could push that higher, that we would just assume that we're gonna get that. So if you look at the revenue miles that we have today, that we're reporting from last quarter, versus the revenue that we actually generated, you see we're actually achieving about $2.25 per mile today. So it's a little bit of a false comparison between those. There is profitability in the truck freight market, as most people in this room know, but it is on average, when you look across the entire industry, it's a fairly slim profit margin, except for the really efficient and really top-notch operators.
We just tried to set the expectations for our revenue per mile in line with what we're actually achieving on the routes that'll be the first to commercialize.
Thanks for that clarification. It seems to be like about a four-year push out of your targets from your original IPO targets, but a two-year commercial delay, right? If I think about what Xiaodi talked about, pushing production out to 25, maybe 26 on your commercial production. What has shifted? First of all, what's shifted in the six months to make such a dramatic extension of the targets? And do you view this as more breathing room for your targets? Is there, you know, is there some scalable potential in that maybe? How do you think about that?
Well, I mean, I think you have to think about the context on when we initially struck the joint development agreement with Navistar back in 2020. This was a very robust test that the TuSimple and Navistar teams went through together to set a start of production for 2024. You think about what has happened from the midpoint of 2020 to the midpoint of 2022, and it's sort of unprecedented in recent times in terms of the supply chain disruption that has happened over that time period. The supply chain disruption, it's not a direct effect on our particular program, where we're waiting for a particular chip or, you know, some particular natural resource to come through and enter a factory.
All of the dislocation and disruption that's happened in the supply base has meant that future projects, like building a component specifically designed for L4 trucking, all of those timelines have gotten disrupted, and there's a knock-on effect as each individual component's timeline gets pushed out, that has a cascading impact on the overall production timeline of the truck. We don't like the fact that we have a delay in the production timeline, of course. We, you know, we set targets, and we set out to achieve them, and we've been, you know, we think that's our calling card, is hitting our milestones. We also, more than anything, wanna be transparent about that. We have pushed out, you know, the timeline by a year or so.
I would say, you know, in terms of the revenue targets that we have, these are really high conviction revenue targets for us, ones that we think there's a, you know, multiple paths, multiple degrees of flexibility to be able to go out and achieve those. We thought it was the appropriate time here, approximately a year after our IPO to go out and level set and provide some expectations to the market.
What I was gonna ask you during your prior presentation, right, just to clarify that mix, right? Carrier-owned capacity now is 2026. The commercial availability for TuSimple only, you're gonna take those trucks in pre-production, run those in 2025. Is that? Then China, I think you said was 2025 also. Is that?
Yeah. That's the general way to think about it. Starting with the U.S., we will start to ramp with the pre-production prototypes starting in 2024 and into 2025. As we get to those sellable vehicles, I mean, that's a critical point of distinction, is that we don't see any way to put the trucks into the carrier's hands in any type of scale until we have something that is certified and warrantied by an OEM. We think that's really important to actually transfer the ownership of the truck to a third party. They'll obviously work very closely with us operationally for the virtual driver service, but that's an important point where we actually unlock that carrier-owned capacity business model. We talked about we wanna be the first users of the technology, and we have to because it is such an advanced technology.
That's why that TuSimple capacity ramps in 2024 and 2025. That'll be with production intent prototypes and maybe some of the early version of the sellable, scalable trucks. Very quickly, as we go through the second half of the decade, as we've said all along, our intent is for the carrier-owned capacity to be the vast majority of the trucks that come out onto our network. We think that allows us to serve the ecosystem better. It's a relatively high margin, capital-light approach, and we think it works best to get as many trucks out under AFN, which is our ultimate strategic ambition.
Are there 10 prototype trucks or 10 different models of prototype trucks? What's the 10 prototypes that you're-
I'm not sure if you're referring to. If you're looking at the scale-
No, no. I'm sorry. I thought you said there were 10 prototype trucks.
No.
No?
No, no. I don't know what that one was, but.
You're gonna start providing number of trucks. By region and thing on the quarterly?
We do have. You know, I think we've been pretty transparent about our fleet size. We put it every quarter where we are. We've got about 100 trucks that are in our fleet, and we've said that our driver out fleet is approximately 10 trucks that we operate today. As we scale those operations throughout 2022 and 2023, for the driver out specific fleet, we'll provide an update. As that fleet grows from 10 trucks to 15 trucks to 20 trucks, we'll provide an update on the subset of our fleet that is really driver out capable, as well as the other part of our fleet, which is driver and AV capable.
Hey, Brian Ossenbeck from JP Morgan. Just to follow up on that last one, Pat, what type of stats or learnings are you gonna provide from the driver out fleet as you start to ramp up more and more miles? Are you gonna have a white paper on it, or how are we gonna see some progress with that as you ramp and build off of that to the ultimate commercial model?
Yeah. I mean, I guess you're probably asking about like specific metrics on our driver out operations. Look, I think we'll provide, you know, very specific examples of things we did. In this quarter, we provided, you know, we said that we did the MRC, which I think is a. You know, if you look across the industry, this is something that is really critical to be able to have a scalable and efficient operation. We showed you the video today. I think we laid out very specific milestones, removing this, removing the support vehicles, adding a new route, being able to do daytime and nighttime operations, and we said we would do all of these routes, including expanding into Texas. We said we would do all of these things by the end of 2023.
For us, yeah, we'll provide some context about how our operations are scaling, but it's not really just about, hey, we've done this many miles, 'cause we could probably generate the most miles just running back and forth from Tucson to Casa Grande every, you know, every night and rack up miles, but it wouldn't actually be making progress to commercialization. These milestones, which help us drive that operational cost down and will provide the operational cost estimate at the end of the year, that's really the way that we would measure progress. It's very easy, right? We'll show the uncut video on how we're actually making progress on that. You know, we'll show that the operations are expanding and so.
That's how I would say we would ask you to measure the progress around that, and then you can sort of have this operating cost per mile is the kind of all-encompassing, are you actually getting efficient? Are you driving towards real commercialization?
Within that stair step down, there's obviously the ADC. Xiaodi, can you talk a little bit more about the costs and the risks associated with that? I know you decided to bring it in-house, but is there a risk that that can slip further out to the right? Do you feel comfortable, you know, with all the partners you have right now? 'Cause clearly that seems to be something you're focused on to get this whole thing scaled and safe.
Yeah. I think the project is still in the early stage, so I can't really comment too much. Of course, there's some non-disclosure terms from with NVIDIA. We're in the early stage. TuSimple is building, for example, the sensor interface and also the safety and redundancy periphery circuits, and NVIDIA is basically providing the reference design and also the chip. That is the thing. We're in the early stage, but we're ahead of schedule, in fact.
Last one from me. Just going back to the $1.50-$2.50 comparison, I know there's some things that aren't in there. Can you just give us a high level of, you know, where you draw the line in terms of the development versus not? 'Cause I think that'll make a difference. Then also, when you think about practically speaking, I think you're excluding some drayage and some terminal costs and other things like that. So, you know, is there a risk that the drayage you're gonna be seeing different rates throughout the cycle, but-
Yeah.
Yeah, I guess, how do you think about the drayage component of it? Because that's clearly, you know, part of the all-in cost here. Just making sure we're comparing this on an apples-to-apples basis. Thanks.
Yeah. I mean, I think the drayage cost is a good one, and it's hard to figure out how to operate that or to incorporate that entirely. I think that's a totally fair point. I mean, I would just point to the fact that what we're trying to do in terms of the AFN is try and connect as much directly to our customers' terminals. You know, our customers don't wanna turn a one-legged move into a three-legged move, where you have to do, you know, drayage to our terminal, we run it autonomously, and then there's drayage on the back end. We are intending to make these kind of point-to-point moves that wouldn't in, you know, take drayage.
That's hard, and it requires a lot of work from Xiaodi and the team on being able to figure out surface streets and get into, you know, industrial parks that are close to the highway, but still could be 1 mi-5 mi of surface streets on either end. That point, there is no drayage incorporated into that operating cost. As we start to ramp up to the point I was making, and we do more and more runs, and we know there is a drayage component, we'll start to work on how we incorporate that, what percentage of runs require drayage, and do our best on trying to incorporate that cost in. The point about terminals, I think it was important that we set the baseline on thinking about what the human operated cost is when you take ATRI.
We haven't incorporated the terminal cost. When you think about like the update from where we were at the time of the IPO to where we are today, one of the things that we're really pleasantly surprised about is the way to be very capital efficient in the how we build out the terminal network. Things like working with Ryder and being able to add two new terminals in Texas that already have things like wash bays and fueling stations and maintenance operations there and doing that in a very capital light way. As Jim was mentioning, we have a couple other partners that are working with us around new terminals that will be co-located at their freight location.
We'll have our own dedicated space, which we'll need initially, especially for, you know, trying to develop the driver out operations. Being co-located really reduces that drayage opportunity. We think there's, you know, from a terminal cost perspective, it's actually gonna be, you know, at least from where we stand today, we think it'll actually be a lot more capital efficient than what we thought it would be even at the time of the IPO.
I think I just wanted to add one more thing is that I think our technology is really having a very deep collaboration with the commercialization side. That is a joint project to optimize. We always have that in mind in optimizing the total cost.
Great. Todd Fowler with KeyBanc Capital Markets. On the removal of the support vehicles, it seems like that's a big component to bringing the cost down here in the near term, and it sounds like the advanced MRC is a big piece of that. What other pieces do you need to see to fully remove the support vehicles here in the near term?
Well, I actually already said that in the presentation. One is that to remove the lead vehicle, we need to have collaborative mapping and then like a vehicle to cloud to fleet kind of communication so that every vehicle is aware of the situation. We actually already have that. It's actually shown in some of the demo runs today. The next step is to further polish that to remove the lead vehicle. That's probably what will come sooner than the second part, which is removing the chase van. Because removing the chase van requires us to have more ruggedized hardware. For example, if some of the components that we're using are prototypes, and they just break very often, several thousands of miles, there's a breakage. We just can't remove the chase van because otherwise the rescue costs will be too high.
We're actually migrating from some of the A sample, early samples to B samples, which are much more ruggedized samples. Once that is complete, the hardware is complete. The second part, which I already show MRC, like stopping at the roadside. Then we also need a centralized, operating, like, tele oversight center so that we can locate every vehicle. We know where they are, so to better help coordinating the rescue service whenever we need.
Good. It sounds like the visibility into that's pretty strong at this point. Just for a follow-up, and I appreciate the update on the purpose-built vehicle, and obviously, if you're just putting out targets, you know, you have some visibility and some confidence in that. You know, it's an important piece to get to the scale that you need. Can you speak a little bit just to the visibility into 2025 and maybe Navistar's commitment and the effort that they would put along to make sure that that can happen at this point? And also any commentary on the cost differential between a purpose-built vehicle and a current commercial truck.
Yeah. I would say maybe I'll just make one comment. It's everything that we put around this production program has gone through very extensive vetting with TRATON GROUP. We are. This isn't a TuSimple opinion. This is us working together with our partners. You know, as appropriate, right? We really are integrated in doing this together. So I'll just make that comment. I don't know.
Yeah. I think the difference here is, like retrofitted trucks, we can, after a careful design, especially with our awesome hardware team, we can make retrofitted trucks reliable. However, we cannot make retrofitted truck cheap, and we cannot scale up. For cheaper production and for bigger number of production every year, we need to work with OEMs.
Do you have a cost estimate right now on where the truck will be relative to?
Yeah. I mean, I don't think we see any real change from what we expressed during the IPO of probably at scale, at run rate of probably a $50,000 or so incremental CapEx associated with it. If a human-operated truck today is $200,000, around $250,000. Obviously, we've seen truck prices escalate quite a bit over the last 12-18 months as well, just given some of the shortages.
Thank you.
Hi, Dan Levy, Credit Suisse. Thank you. Wanted to start just to ask on the path to revenue. I think in the beginning, you had a slide that showed just the concentration of goods movement within certain routes. I wanna understand, in the revenue targets that you've articulated, what does that assume in terms of the number of routes on which you'll be operational? To reach initial scale, maybe you can give us a sense of, you know, how many routes that requires. Is it enough just to be within a few routes in the Sun Belt? Or at what point to achieve these numbers do you actually have to start to go up north and achieve, you know, greater capability in challenging weather, et cetera?
I'll just say from a financial perspective, we think, you know, for those initial years of commercialization, we're not baking in any ability to expand the ODD into the northern half of the country and be able to do snow. The Sun Belt in and of itself, where, you know, if you ask third-party estimates, it's a $30 billion market just on I-10 alone. There's a tremendous amount of freight on a relatively small number of routes. I don't have, you know, for each one of those years, a specific number of routes that we expect to be running. It's obviously gonna be more than a handful, and it'll expand somewhat proportionally to the revenue targets there. Nothing in that is assuming that we're gonna be running, you know, on I-80 across the northern half of the country.
Okay. Thank you. Second question related. The path to achieving that revenue from a technical capability, is that contingent on any AI or machine learning breakthroughs, or can you do this purely with your capabilities? Again, you know, I think you've talked about some maturity required in the underlying components, but be it steering, braking, LiDAR. You know, maybe you could just give us a sense of the gating items that you need to unlock that scale, and start with the AI piece, please.
Yeah. I think it's probably more of a. The short answer, no technical breakthrough needed for this. Everything's on the engineering side. Devil is in the details. I think the most time-consuming part is actually the validation process. You know, sometimes you like in the auto industry, like you build a car, you put a car to the Furnace Creek in Death Valley, and you also put a car to Alaska to do all the tortures to make sure that the car is validated and verified. It's very similar thing to our technologies. Like, we feel that we have the technology, the technology is great enough, but we still have to do tons of validation, probably 10 times or even more than a lifetime driving of any average human driver like you and me.
Only after that we see first of all, there has to be no safety issue. Second, all of the efficiency-related issues like number of harsh brakes, number of braking and leading to the minimum risk condition. They're all within the control so that they make business sense. Only when these two things are achieved by validation, which is very long but necessary process, we can claim, okay, so this route is driver-out capable and that route is driver-out capable.
No, no AI needed, correct?
Small tweaks, but no major AI breakthrough needed.
Okay. Thank you.
All about engineering.
Last question. I appreciate the financial update and the EBITDA margin target by 2030. Maybe you can give us a sense of maybe what the rough timing is to get to EBITDA breakeven, free cash breakeven, what the flow of the underlying R&D and SG&A is.
Yeah.
is gonna be. Thank you.
Yeah. I would say if you start to think about, you know, we talked about the unit economics, and you obviously think about what's that run rate cost of ongoing R&D and corporate costs that we need to keep it going. Say it'll continue to expand, but at a much slower trajectory than what we've seen from 2020 to 2021 to 2022. I'll ask Xiaodi to comment on this, but I think it's, we've really hit a critical mass in terms of the team that we have here. There's parts that will continue to expand, but you shouldn't expect to see that the R&D cost is gonna balloon to be, you know, significantly higher than where we stand today in order to achieve commercialization.
The other part, because we are working with OEM partners for manufacturing, there's also not some big CapEx cliff that's happening over the course of the next couple of years to build a production plant or anything like that. We are a technology company, not a manufacturer. So you don't have some of these big escalating costs, where you think about that as a real drain on your margin structure. In terms of where we get to from, like an EBITDA or cash flow breakeven, we're not providing a specific target. I will say that if you look at the unit economics, it does happen fairly quickly as you start to get into the scale of trucks.
Once you get to the, you know, you're starting to measure that in thousands or, you know, 1,000 or multiple thousands, that's when you can start to look at the unit economics, and it starts to cover all of those corporate costs and R&D costs that are gonna be kind of running in the background to support the operations.
I've got one from the webcast here. How do we think of today's competitive landscape and how it's evolved since your funding? Why do you think your business model will prove superior to the existing business models over time?
Should I start?
Mm-hmm.
Okay. Yeah, good question. I think, you know, we should see from the facts, we're the only company that's driver out, and we believe there's tons of things that you need to do for driver out. I believe, I think after today's presentation, we all see that, there's just a lot of things behind the scenes that you don't get credit for a marketing campaign out of this. Those are the hard work that needs to be done, and we're very proud that we have silently finished all of them, and we're ahead of anyone else by a large margin. That's my feel.
Ravi Shanker, Morgan Stanley. Thanks for the follow-ups. A couple of questions for Jim. To the extent that you can comment on a public forum, what is the issue in California, and when do you think that might get resolved in terms of approval for Level 4 operations, if you were to hazard a guess? And also, point well taken on this being federally approved, but you need to resolve some of the operational implementation of this. It would be a shame if you guys figured out how to drive a truck by itself, but you couldn't figure out how to throw a couple of reflectors out. How complex is that situation, and what do you think the timeline for resolving that is?
Sure. I'll start with California, and I've had a history with California, so I don't think I can say anything that would put me any more disfavor than I already have. We're very close with our competitors on these issues. We all have lobbyists, consultants in the state. We have active conversations now. We made some headway recently with the transportation infrastructure chairperson who finally said, "Enough's enough. Labor, you gotta come to the table. You gotta come to the table." We did. I can't say that they're overly enthusiastic about the whole situation, but I think they realize. I mean labor realized, okay, this technology is here, it's going to be adopted into the freight network.
I think as they give it more thought and how it affects their membership, I don't think it's as detrimental as they may have thought before. I think they're working through those issues. Between the legislature and the governor's office, we're making some headway. Now, sometimes with folks, you couple steps forward, then a couple steps back. I'm not saying this is gonna happen in 2022, but I think that you'll see them do something similar to what they're doing in the past cars. I think you'll probably see where, you know, Texas and Arizona, you don't need special permits, special certificates. I suspect you'll see that in California, in the next couple of years. The second question again, Ravi was-
The rule making for the
Yeah, again, associations with our competitors. We have a call tomorrow with the Department of Transportation, both the office of the Secretary's Policy Office with FMCSA with NHTSA. Our agenda is to impress upon them the importance of getting this rulemaking done near term. They've talked about when you go through the rulemaking. There's three steps. They've already started step one. Step two is a notice of proposed rulemaking, and that's where we are trying to get them to now, and they said November. That's okay. November is okay. With the government agency, sometimes November of 2020 means November of 2023. We just need to impress upon them that the technology is here. It's not just us, right?
Our competitors are all in the same but not all, but some are in the same position we're in. We're having those conversations regularly. They're coming down to Tucson in June to do what you guys are doing today, get on the truck, see the technology firsthand. I ran into the chief counsel for FMCSA at a conference last week. His name is Earl. Earl, I chatted, and he says, "Look, Mo, we get it. We understand that you and a few of the folks in your space, you need this to be resolved near term, and we're working on it." The Secretary Buttigieg has made some very positive comments about the technology and the benefits of the technology.
I was having a discussion with the UP the other day, and I talked about Buttigieg's position on autonomy and being very supportive. She said, "Well, not on all modes," but signals are good. Signals are good. Lastly, on the legislative side, sometimes people ask, "Well, what do you need from Congress?" Truthfully, I say this to my friends that are in Congress, I need you to just don't screw it up, right? That's where we really just don't screw it up. There are those that are optimistic that there would be a federal preemptive landscape. I don't see that happening near term. Maybe at some point in time. I don't think that's gonna happen in the near future. All then, the holistic of is things are good. We work very closely with Arizona, Texas.
Florida wants us in there. I think we get calls saying, "Hey, when you coming? When you coming? When you coming?" I don't see any headwinds that are gonna prevent us from hitting our milestones or our timelines.
Got it. One follow-up. I know this is for Pat or Xiaodi . Again, I know you have a lot of work to do, but here's a hypothetical. If one or two other OEMs not named Navistar or TRATON showed up at your door saying, "You are amazing. I want you in my truck," would you engage? If you did, how hard is that process going to be? Is it like starting the clock from Navistar and at the same time from this point forward? Or because you're doing all this work with Navistar and TRATON, you can scale that up in the same timeline you've shown us here. Thank you.
Okay. Yeah. First, as a matter of fact, our collaboration with Navistar is non-exclusive, like, both ways. Also, I wanted to just warn everyone about the level, the depth of this collaboration is very, very deep. We have to take everything very, very seriously. That being said, it's very unlikely for us to just go with everyone and just, you know, collaborate with everyone. It's probably not going to happen. Yeah, this is a very deep collaboration. That's all I can say. Yeah.
Hey, guys. Appreciate the time today. It's Darren Baker from PRIME CAP. Actually, Jim, I think this is mostly a question kind of directed to you. You know, in just talking with folks on my team, I sometimes get the question about, you know, what happens when we get more of these trucks out there. They're not being followed by unmarked law enforcement. They don't have a survey vehicle. They don't have, you know, a chase vehicle. They don't have any of these things. Wherever, you know, whether that's a prototype vehicle or something in production, and we have the first kind of big accident, right? Something goes wrong, and not necessarily with you guys, although perhaps it could be, right? Some non-compliant driver or some just bad thing happens, right?
You know, we're looking at the front page of The New York Times and The Washington Post, and they're talking about, you know, the danger of these unmanned vehicles driving around with, you know, a gross weight of 40 tons. You know, what, how would you guys respond to that, if it was one of yours? If it was not one of yours? You know, and what do you think, do you think there's risk basically that that kind of has the potential to, you know, get Congress involved, to set back the industry because it just creates sort of an image headwind that, you know, right now maybe is not sort of in the public eye, but could arise in a circumstance like that?
Yeah. Yeah. That's obviously a concern that is an issue that all of us in the space have to confront. I guess I'd just first start off with, of course, we have a crisis management team. We have a plan in place when that happens. Our fault, not our fault, it doesn't matter. We have a plan in place. What I always talk about when this question comes up is because it's probably it's gonna happen, right? There's gonna be an accident, and it's gonna be in the news. It's incumbent upon us to promote the benefits of the technologies, so that people have an understanding of all the benefits before that bad episode, that bad event occurs. You talk about the political winds, right?
If the accident happens in, pick a state, Texas, and then all of a sudden Senator Cornyn or Cruz, "Okay, I gotta take a stance." We need to make sure that we have already educated those folks, and we do that all the time. We're on Capitol Hill. We're in the state houses. We're educating folks already on what that looks like. The thing that we don't like to talk about, right, because it pokes the industry in the eye the way it is now, right? We have 5,000 fatalities every year involving commercial motor vehicles. It's been stagnant for years. 30,000 + serious bodily injuries every year. We can't let.
We talk about this with the regulators all the time because there was, originally it was like, okay, before we, quote, "allow" this to happen, how safe is safe enough, right? Do you have to be better than 90% of the human operators, 70%, 50%? Where do you set that measurement? You can't because it doesn't make any sense. Nobody wants to put themselves in that position either. To get back to the crux of it is we have to do a very good job of educating the general public of the safety benefits of this technology, so when that bad event does happen.
It's, you know, well, there is a provider that has some sort of a driver assist system which may or may not call itself driverless. We can't do that, right? We pride ourselves on being very transparent, right? When we did the drive route, we worked very closely with ADOT. We worked very closely with the federal level. That's how we're trying to, again, soften the target with the transparency and promoting the safety benefits of the technology.
Sure. Thank you.
I wish there was a silver bullet. There's not, right? It's the same in the manual space, right? You see this, the litigious environment and these nuclear verdicts. It's a real concern. The kind of a side benefit of this is when we're involved in an accident, our technology is gonna show exactly how that accident happened. There'll be no question about how it happened, where the motorists were, where the accident happened, the relative speed of the vehicles. That's a collateral benefit because we do believe this is gonna be a much safer product. Having the ability, when you try cases, everybody says, "Well, if you only show the jury a video." Well, we will, all right? That'll take some of the heat out of the situation immediately.
I think we're about out of time. We do have a reception two doors down in the grand foyer with management, so you can feel free to ask them more questions. But on behalf of the entire TuSimple team, I just wanted to say thank you for making the trip down. We really appreciate it. We really appreciate your time, and we're glad we got to tell our story. With that, thank you.