TuSimple CEO Cheng Lu sketches the promising road ahead for driverless trucks

In December, self-driving truck startup TuSimple joined the list of autonomous big-rig companies reaching driverless milestones.

One of the company's Class 8 trucks completed an 80-mile ride between a rail yard in Tucson, Ariz., and a distribution hub in Phoenix without a human in the vehicle. The driverless trip took about an hour and 20 minutes, on surface streets and highways, with the truck making lane changes in an otherwise normal traffic environment.

With more extensive driverless testing scheduled in 2022, and the self-driving truck industry gaining traction overall, TuSimple CEO Cheng Lu talked with Staff Reporter Pete Bigelow during CES about the company's "driver out" development plans, ongoing work with Nvidia on a "system on a chip" designed for trucking, and TuSimple's burgeoning autonomous freight network.

Here are edited excerpts.

Q: One of the most interesting things in the 80- mile ride is that your truck ran on both surface streets and the highway. Obviously, you have competitors who think operating on surface streets is not feasible and are pursuing a highway-only model. How are you doing on surface streets now, and how much more challenging is that?

A: It's a very different feature set we're solving for, because you have different edge cases. You have more pedestrians and bicyclists. You have unprotected left turns, which is the first thing we do when we turn out of our Tucson facility. That means you have to capture cars coming at you at 30, 40, 50 mph. That's about 20 meters per second of velocity. Therefore, if you don't have the perception to go past 300 to 400 meters, every car coming at you, it's a false positive. Your truck would never move because it takes so long to make a left turn. That's very different than a car. So this is a totally new game, some of the things we're solving for on surface streets.

Why do that? Why go for the harder problem instead of a highway-only model?

Technology adoption works the best — you see this over and over again — when we're catering to what the consumers want rather than making the customers change their model to fit the technology. We talk about this all the time. You're not going to see a large carrier or shipper go to a highway transfer hub. It sounds cool, but it doesn't happen.

 You've been developing this technology for years. How do you know on Dec. 22 that you're finally ready to do this driverless run? How do you know it's safe?

We embarked on taking the necessary advancements both on hardware and software to get to driver out over the last two years.

Once we had a very clear validation and passing criteria, that's when we felt comfortable. We also, prior to that, published our safety case, which was about 50 pages. That's the preamble, actually, because we can't publish the proof.

A lot of that is confidential and sensitive information. But we need to have a comprehensive safety case that leads the industry. So it was really two years of hard work, a very comprehensive safety case, a comprehensive validation plan to feel confident about the driver out. 

In addition to developing the self-driving system itself, TuSimple is building an autonomous freight network of roads where you can operate. How are you growing that network, and why is that such a priority now?

Of course, the core technology is what everyone is focused on. At the same time, we're building this in parallel. One analogy I like to use is if you wanted to provide a 5G telecom service, you wouldn't wait until the 5G phone is built before building cellphone towers. That would add another year or two or whatever it is. So that's how we think about it. We get a network in place and get customers accustomed to autonomous freight. And when the technology is scalable, then it's plug and play.

Let's talk big picture a minute. Obviously, you started pursuing this before supply-chain problems were the headline every day. What has the newfound focus and appreciation of supply chain and logistics done for autonomous trucking? And has it generated more interest from truck operators?

The hiring of drivers has become very difficult, and it really always was. But I'd say now the driver shortage has become a mainstream thing. Most people, when they pass trucks, you don't really think about how trucking works and you don't think about the 2.5 million Class 8 trucks that power the supply chain. Now I think people definitely understand. Or at least feel the impact. So I think the tailwinds are very strong for what we need to do.

During CES, TuSimple said that it's using Nvidia's Drive Orin system on a chip as the basis for a new partnership, where you're joining forces to develop an autonomous domain controller specifically for TuSimple and autonomous trucking. What does this bring to your skill set that you didn't have before?

You need a central computer that is very powerful to run all our algorithms.

You need something that is a small package and auto grade, and something that has certain levels of power consumption. Today, there isn't something like this in the market. You can draw a parallel to the EV space, where I think the early leaders have to be more vertically integrated.

If you look at Tesla — and we're not comparing ourselves to Tesla — you start to understand your power system and your BMS [battery management system] and your own chips. It pays off, because it allows you to be first to market and better control your supply chain. I think the industry rewards these kinds of moves.