What Will It Take to Make Autonomy Real at Scale?

本站添加: 2026-05-04

[00:00:02][music] >> Okay, hello. Thank you for being here.

[00:00:10]Yeah, thanks for having me. Okay, so Chris, you've been in the autonomy game for a long time.

[00:00:16]And we've been hearing that autonomy is almost here for over a decade. Yep. Uh you're now running Aurora's now running commercial driverless trucks. Are we at the start of scale or is this just a polished pilot phase?

[00:00:30]We are very much at the start of scale.

[00:00:32]Uh so it it it's been kind of an awesome experience uh living the Gartner hype curve for the last 15 years.

[00:00:40]>> [laughter] >> Uh you know, and and there's this kind of period with the Gartner hype curve where everybody's like, "Oh yeah, it's going to be here tomorrow and magic will happen." And we had that experience in 2015, 2016 and uh and then unsurprisingly, we spent our time through the sea of disillusionment or valley of disillusionment and and now we're very much at the other side of it.

[00:01:00]And for Aurora, this is really exciting.

[00:01:02]We've got a handful of trucks we're running driverlessly today.

[00:01:05]Uh and by the end of the year, we expect to have a couple of hundred of them running across the southern US. So, big big step for us. Yeah, I mean you say that we're out of this hype cycle that I think that we've been in. But is that a good thing? Do you need the hype to get the money to fund this extremely expensive technology? So, I I'd say that we are in a really good position. We have, you know, a lot of money on our balance sheets and we're well positioned to go build this company. Um and for us, I think I would much rather be in a world where people are seeing and realizing the value of what we're doing rather than listening to more promises.

[00:01:42]Uh and that's what we're starting to see with customers. We have folks like Hirschbach, uh Werner, others that use our trucks day in and day out and they get it uh and they want to see it scale. Mhm.

[00:01:53]Now, Aurora made headlines, you know, last year when you became the first, I will say, kind of to commercially deploy fully autonomous trucks. But there is still a human in the cabin and you have your reasons for that. Does it still count? We work very hard. So, so yes, let me answer your question directly. Yes, it still counts.

[00:02:14]Uh we did 6,000 plus miles with nobody in the vehicle. Um surely without somebody behind the steering wheel.

[00:02:22]Uh for for uh partner's request reason, we've been operating for the last year in a driverless mode with an observer uh in the vehicle. And we hold the the standard and the rigor with which we release the software in the system to the same bar we would as if there was not somebody there.

[00:02:37]Uh and that's really important because uh in Q2 of this year, we expect to be operating our new fleet of trucks that the vehicles are International LT trucks and they'll have our overlay of by wire and our drive by wire capability or sorry, our second generation hardware capability.

[00:02:53]And they're going to be operating with nobody behind the wheel. Uh and so it's been great to have this period of time to build that experience. We're well over uh 250,000 driverless miles to this point. Uh and then to be ready to go and scale through the course of this year.

[00:03:07]Mhm. Okay. Now, I want to talk about some of the bottlenecks. So, you know, today you can run your autonomous trucks in Texas, but not California because of California's, you know, state-wide restrictions on heavy-duty autonomous vehicles. Um if California How much of your road map is dictated by regulation versus technology? So, I guess I guess in the strictest sense, our road map is dictated. We don't do things that are illegal. Uh and so if California says we can't run in California, then we wouldn't run here. That said, the vast majority of US states already allow us to operate and we can build a hell of a business just with the states that are open today. I think it's something like 50 billion VMT miles that we're going to be kind of opening up over the next several you know, next year-ish. Uh and that's the Sunbelt without California. What's exciting though is that in the states that we operate in, in Texas, Arizona, New Mexico, people are excited about the technology. And we've been working closely with uh the regulators and uh executive branch and the legislative branch in California.

[00:04:12]And they've been going through a regulatory process to open up California trucking. And so we're excited to see that. We're expecting the regulations to come out in the next month or so. Uh which would unlock California for trucking and see the state benefit from the safety, the sustainability, and the economics that'll come along with this.

[00:04:30]Okay, well, I mean so if California opens up, does that unlock real scale or are there still some technical and economic barriers? Yeah, so California unlocking will be great, but again, we'll build a business with or without it. Uh and ultimately, it's up to the people who represent Californians to determine what the policy is for this state and like any other state. Um for us, really the the constraint has been our ability to supply the vehicles. So, the first generation hardware we have and we've been very clear about this, we could only build a few of those. And so we've been operating with them, we've been learning from them. Our second generation of hardware will unlock and will launch in Q2.

[00:05:09]And with that, we can build 1,500 units-ish. Uh and then what's really exciting is in 2027, uh the hardware we've been co-developing with Mobileye, which is uh the unit that spun out of Continental, which is one of the world's leading tier one manufacturers, that hardware will come online and that we can produce tens of thousands of units of. So, for us, we've had this opportunity to pilot, to learn with customers, to help them understand the value to them. We're now through the course of this year going to go from a handful to hundreds of trucks with this new hardware that'll build. And then next year, we'll be in a great position to truly take this to uh to very large scale. Yeah, and so that Continental partnership is really crucial for you scaling and getting that um you know, hardware uh up to snuff. So, you're expanding routes this year. You've got 200 trucks on the road today. I think by the >> a handful of trucks today, 200 by the end of the year.

[00:05:59]>> 200 by the end of the year. I think you want to scale to what, 2,000 by next year?

[00:06:03]>> Uh by the end of next year, yeah, we'd love to be in a couple of thousand.

[00:06:05]That's wow. So, okay, so when it comes to getting um chips for our onward inference, are you fighting against AI companies and hyperscalers for parts today? Uh not really, no. Uh at least not directly. So, we use uh Thor SOCs.

[00:06:21]At least that'll be what we put into our next generation uh Mobileye hardware kit. And so that's just a different um a different chip. Uh it's targeted for automotive and physically AI applications and we'll be one of the lead customers for that. Okay.

[00:06:35]Now, tying it again to the hype cycle.

[00:06:38]So, autonomy's had this brutal hype cycle. Robotaxis were supposed to be everywhere by now. I mean, they are starting to be.

[00:06:45]Uh what is different now and what is different about trucking?

[00:06:51]Yeah, so what's different now is we've had a good chunk of time to actually do the work.

[00:06:56]Uh and to get to a point where we can responsibly and safely bring this technology to market and actually have a product that people want to use. Anyone here who's had a chance to ride in a Waymo, it's kind of exceptional. It drives really well. You You know, if you stand on a corner outside here, you'll see two or three or four of them go by.

[00:07:13]>> Oh my god.

[00:07:13]>> Uh it's super impressive. Uh I'm really proud of what that team has done and and seeing that come to market. With trucking, we're very excited about this as a company. Uh it's a place where we see real need. Uh the US economy runs on trucking. Uh everyone who's here should you know, if you know a truck driver, please say thank you. It's a really difficult and important job that not enough people want to do.

[00:07:37]And the technology we're bringing to market will allow us to complement people who drive trucks. Uh allow us to do that where we expect to save fuel, 14 to 34%, which is great for the environment, great for the companies.

[00:07:49]And allow the companies that that are trucking companies to basically double the utilization of this really expensive asset that they buy. And so for us, given we have the technology to go and solve that problem, it's a really interesting business to go build. So, okay.

[00:08:05]That's interesting that you say that.

[00:08:06]So, you know, you and you mentioned Waymo.

[00:08:09]Waymo's uh you and Waymo kind of started in the same place. You were both pursuing trucking and robotaxis at the same time. Waymo ended up, you know, ditching Waymo Via um and focusing all all on robotaxis.

[00:08:22]You did the opposite. You ditched the robotaxi or at least put it on pause.

[00:08:25]Yep. Focus more on autonomous vehicles.

[00:08:27]Is that is that a technological decision? Like it makes more sense to in a to do autonomy in a constrained repetitive environment or was it like the business opportunity seemed larger for trucking? So, it's a combination of things. So, first, I think it's really important to understand that uh driving a vehicle, particularly a big truck down the freeway at 70 mph, is not easier than driving in an urban environment. It's just differently hard.

[00:08:52]Okay. Um right? The fact that we have to look, you know, a half kilometer down the road, the fact that, you know, the kinetic energy is so much even with this thing and the reaction times and the distance we need to react things. It's just a different hard problem.

[00:09:05]For us, there's a two reasons why ca- primary reasons why we focus on trucking. So, first is we do have a technology we think that is unique that allows us to solve the problem. It's a really hard thing to see far enough down the road that you can drive safely with a big truck. And we have been developing this technology called First Light, which is a special kind of lidar that can due to kind of the basic physics of how it works, can see dramatically further than uh conventional lidar can.

[00:09:33]And that's a big deal when you're talking about uh again, a 70,000 lb thing moving down the freeway at 70 mph.

[00:09:39]Uh the other part of it, so one is this technological advantage and we've done a lot of really exciting things around that technology and with verifiable AI.

[00:09:47]But the other part is uh the business.

[00:09:51]That the economics of trucking are large. So, it's a trillion-dollar market. Ride-hailing is about a $50 billion market. So we can see the opportunity to grow the business without having to chase down cost. Uh and then the unit economics are greater that we value a truck being driven at three or four times what we each value driving an Uber. And so as a company bringing new technology to market, having a gigantic market, having stronger unit economics that mean when you're entering you could be profitable sooner, and then you know, having the huge opportunity to reduce the 5,000 vehicle deaths that occur related to trucks in the US, to improve fuel economy by 14 to 34% to you know, double utilization for customers. That all just really makes it a unique and exciting business for us to go build.

[00:10:39]Okay. Well, you know, we're talking commercialization, but and good unit economics, but revenue is still pretty minimal, right? Correct me if I'm wrong, 2025 we had uh Aurora had 3 million on the books for revenue. That's on a $900 million burn. So when does this stop being an R&D project and when does it start being a real business? Yeah, so it is a business that's growing in an exciting way right now. So you're right, the the last couple of years they've been the the revenue we've generated has been I think the finance people call it de minimis. Uh but it's allowed our customers to actually understand the value of this technology for them and allow us to build the demand that we see.

[00:11:18]By the end of this year we expect to end the year with an $80 million revenue run rate, which is exciting. Uh and we expect to actually become profitable on a run rate basis in '28. Uh and so we see this you know, as you pointed out, we expect to go from a handful of trucks to a couple of hundred trucks to a couple of thousand trucks.

[00:11:38]And when I look at the opportunity for our customers, these trucking companies are really very TCO total cost of ownership focused. And we will help them transform their businesses and it'll in within 5 years if you're not using our technology, I expect you will not be competitive as a trucking company. And so I think this is going to be a very exciting few years for us and for the the freight logistics industry. Okay. I want to zoom out a little bit. So you've spent your career taking AI from the lab to the real world. What are some of the biggest lessons from autonomy that can apply to physical AI which we're all hearing so much about?

[00:12:10]>> Yeah. Well, the good like when I look at what we're doing at Aurora, it is the vanguard of physical AI, right? It is we are driving a big giant you know, robot truck thing down the road at speed. Uh and so I think one of the things that we have put a lot of energy into is understanding how do we know this is going to be safe?

[00:12:30]Now this is one of the big gaps between a chatbot and a giant robot, whether it's a car or a truck or a humanoid or an aircraft, is that there are real consequences that come along with making mistakes uh in the physical world that you don't have in the chatbot world.

[00:12:48]>> Debatable.

[00:12:49]There is some serious mental health concerns. It's just >> yeah, I I I don't mean to diminish those and you're absolutely right. Uh but there's and maybe that points that this stuff will actually traverse more in that direction as we start to understand the consequences and see those uh those models be used in more kind of uh directly impactful applications. Yeah, well, I mean talking about safety, so what does proving safety before deployment mean in practice? Like when do you get to a point of deciding and who gets to decide that the robot is safer than the human? Yeah, so it's for us we have put a lot of energy into a process we call building a safety case.

[00:13:29]And a safety case is this explanation for why we think that the vehicle's going to be safe on the road. And it really um covers five pillars. So the first is that the vehicle's proficient.

[00:13:39]So it drives well. The second is that it fails safe. So if something breaks, it knows about it and then it responds in a way that that keeps everyone safe around it. Uh that it's resilient. So that if someone misuses it or tries to uh make a cyber attack on it, we're robust to that. That the organization is continuously improving. So we learn from what others mistakes others make and mistakes we make to make the system better. And then ultimately that we're trustworthy, that we engage appropriately with regulators. And so those kind of five pillars blow out into 400 and some kind of bits of evidence that we gather, ultimately getting us to a point where we have conviction that we're not creating unreasonable risk on the roadway. Uh and so that's the way we approach it and we've been transparent.

[00:14:21]We are an industry leader in sharing that process. Uh and we engage very actively with the regulators at the federal level and the state level, helping keep them informed of how we're thinking about this, how we're moving forward.

[00:14:34]Yeah. I'm I'm curious, you know, we're obviously uh in there's Waymos everywhere as we've pointed out and I think it could just be the type of people that are here, but every time someone brings up Waymo, the first thing that comes out of your mouth is oh my god, I love Waymo. It's so great. It's so fun. Um has Waymo improved the brand of self-driving at large? I think that we're very early in the experience that we as society have with this kind of physical AI uh and with things like Waymo. And I think anytime we get a win for any of these companies that are doing it responsibly and getting it out in the world and allowing people to move past the hype and the theory to the practical benefit, I think that's a huge win. And so yes, every time Waymo has a success, I am their number one fan and cheering them on. Yeah.

[00:15:23]Um you know, obviously uh AI has existed for a long time, but there it means something different since the chat GPT moment. I'm curious how your business or the way that you do physical AI has changed over the last few years, if at all. It it's continued to evolve. So when we started Aurora 9 and 1/2 years ago, we had a few insights that still hold true today. One is that the system should be learning and that you should be minimizing the places you hand code and trying to draw from high-quality data to to understand how to get the system behave well. And that's been the core philosophy we've had. Our ability to execute that and integrate um larger and larger quantities of data has improved and increased over time.

[00:16:08]But that fundamental underpinning and approach has been core to how we've we've solved the problem.

[00:16:14]The other big insight we've had is around this concept of verifiable AI that we don't believe in having a black box where data goes in like and steering and braking gas pedal comes out the other side. We actually want to decompose the problem so that we can test it and have conviction that this really is working in a way that we would want and and the public would want on the roadways. Mm. Okay. So it's not um necessarily like an end-to-end system.

[00:16:40]No, we we certainly in the sense of thing in black amorphous blob, thing out, we're definitely not that. And you think this is better for safety or We think it's a really important part of how we get conviction that it actually works because it's easy to look at a system and get some experience with it and it seems to work. And we want to do the hard work to convince ourselves that we know that it works.

[00:17:02]Uh and that means being able to introspect and really understand what's happening uh in you know, what's it seeing in the world? How's it responding to that? And making sure we we have that ability. And then also putting guardrails and constraints about how it can behave in ways that are not mutable.

[00:17:17]Um you know, it's no good to kind of tell it that it shouldn't do this because at some point in the future it may determine that it it wants to.

[00:17:24]Whereas if we can actually constrain it, uh that really solves the problem. Okay.

[00:17:29]Is um are you is Aurora all in on trucking or are you trying to expand, you know, using these learnings and expand into robo-taxi soon and then maybe I don't know, just robotics generally or any other So so both are true. So we are very much all in on trucking and that is the business we're building right now.

[00:17:46]But as I mentioned earlier, we see ourselves as on the vanguard of physical AI. And that ability to build a safe thing that can drive a 70,000 lb truck could be driving uh a mining truck or it could be driving a combine harvester or it could be driving a robo-taxi or a box truck or a a walking robot. And so we're going to focus on trucking, we're going to build that business, serve those customers, and then we're going to look at that capability we've built as a company and figure out where the right places for us to go apply that next to help you know, help the world get better. Awesome.

[00:18:18]Well, thank you so much for joining us today. Thanks for having me. Okay. It's fun.

[00:18:23]>> [music]