00:10:09 · Ruby Justice Thelot, Brandon Basso
[00:10:09] S1: a light of a driver. So, honestly, I look at these scenes. I'm like, thank god I'm not driving in London or Japan right now. Well, [00:10:16] S0: this is this is why we say that the the driver should be better than better than humans because I I was actually looking at this image on the left. I'm like, do I do I... Should I know what to do here? Do I got it right The point is that construction scenes are all a little bit different. You have to understand language. You have to understand that language both from a system one thinking level, reactionary, but also a system two thinking level, which is, like, long term strategy. So, like, it's not just enough to parse the sign, you but also need to understand, like, what are the consequences of par... Parsing that sign? How do I make sure I'm not driving over here into this, like, left region? And the... And then if you look at the right, you know, there's there's signs in in in different characters. There's a person there that's like, maybe it doesn't even look like he's necessarily trying to tell you what to do. So this is an example of the diversity of data that we have, and it's... The point is that if we learn how to drive in situations like this in Japan on the right and this in The UK on the left, we're gonna make a generally better driver that works worldwide. [00:11:19] S1: How do you gather all of this data?
00:11:19 · Ruby Justice Thelot, Brandon Basso
[00:11:19] S1: How do you gather all of this data? [00:11:22] S0: We have a fantastic operations team that thinks about that night and day, twenty four seven, three sixty five globally. So that's pillar one is our operations team. Now, Wayve is decidedly not a hardware and operations business. That's a... It's a very hard business to be in. People who have been working on self driving, especially Robotaxi forever, have massive operations footprints. You know, kudos to them for how they've built that out, but that's not how Wayve is planning on solving the problem. We're planning on solving the problem by having a minimal internal operations footprint purely for the purpose of geographic and vehicle diversity data collection and also high fidelity human reference data. So we have expert drivers. We see how they drive. However, we wanna primarily generate data with simulation over time. Mhmm. Maybe not primarily, but over time. That's one aspect of getting data. Two is partnering with our OEMs who have a much richer set of data than we do, and three is purchasing data. Yeah. So there's a rich set of data out there that we can go license or purchase from a variety of vendors. Dashcam data, for example, we have a huge corpus of crazy stuff that's happened on highways. Man. So if we can get that, we can incorporate, something that happened on highway ten years ago into our corpus, then we can build a better driver. So multipronged. [00:12:41] S1: Alright. With that said, I do wanna see the other demos that you have on here.
00:12:46 · Brandon Basso
[00:12:46] S0: Yes. So we kinda jumped into the part that's near and dear to my heart, which is the validation part, how we prove safety, how we think about data in different continents. But I wanna touch briefly on the AI driver itself. And the main message here is that the way that we've built the driving stack is with safety as an intrinsic property versus an extrinsic bolt on property. Lots of systems today are either hybrid or compound AV, or they're purely modular... Sorry, AI, not AV, where they have different subsystems that are responsible for different aspects of the task. And one feature of these systems, it turns out, is there's lots of checking. Yeah. The The canonical example I'll give you, and then I'll come to this example that's on the screen, is maps. Lots of systems have high definition maps that are really like checkers. Like, they say, well, look. We think that we've mapped this before. There's a stop sign right here. We really need to check that there's a stop sign here and stop for it. And there's all sorts of challenges you can get into with systems like that. Like, what if the stop stop sign isn't there? What if it moved? What if there's a temporary stop sign? What if the stop sign is connected to a bus that's moving? Mhmm. And you see all of these different failure modes, and the the reality is that the checker systems tend to be the ultimate bottleneck and limit of system performance. It's not impossible. There are some fantastic systems out there today that you can get in, and we can go take a we can go take a ride right now, you know, hundreds of miles around the Bay Area on on systems like these that are just absolutely fantastic. They were built at great expense, and they require great operational overhead to maintain. And like those systems, but especially Wayve, we're working on making that a much more sustainable system. So net net, we try to learn intrinsically what is safe behavior. What you see on the screen here is representation, a learned representation of safety. You can think of this as, like, a future occupancy map. The red areas are the the don't drive in areas and then sort of a gradient. And you might have seen representations like this before, those familiar with AVs or robotics where it's just sort of looks like an occupancy map, a future occupancy map.
00:15:02 · Brandon Basso, Ruby Justice Thelot
But, really, what this representation is saying is predicting where is and isn't safe to drive based on... You see this bus coming up on the right. Right? So that's a rear a rear bus coming up fast, and it's telling you in the future, it's not going to be safe or it's gonna be marginally less safe to do a lane change into this lane. So we use this system not as a bolt on checker, but as an explicit head, a task that we learn on how to drive safely, and we use this to to guide our ultimate driving trajectories. Now we, of course, have all of the lower level systems that you might expect from a car today that you need to do to be compliant. These active safety systems, automatic emergency braking, lane keeping assist, emergency lane keeping assist is a soup of, you know, TLAs, three letter acronyms I won't bore you with, but the main AI driver learns safety as an intrinsic property. So here's a representation. [00:15:56] S1: Alright. I have some larger macro questions I wanted to ask just about autonomous driving as a whole. This is an area that I'm particularly excited about. If you look at just the incidence of people who die from car crashes and thinking about, hey, maybe in the next generation, that number could in theory be zero. That is insane. And so if we could just zoom out, let's talk a little bit about autonomous driving in general. What does that enable for society at large?
00:16:22 · Brandon Basso
[00:16:22] S0: So many things. And to to just emphasize your point more, those those stats are always very salient. They always hit me hard, especially as someone who has young kids, but they disproportionately affect people who are younger, who are less experienced drivers, disproportionately affects different people in different ways throughout the globe. And I think it's really incumbent upon us to to think very deeply about how we upset the status quo here because the status quo isn't great. But beyond the safety argument, which I think most people, like, sort of intuitively get, to your question, like, what what is... What's different about society? What changes about society? And I think there's gonna be a big shift in 2027 into 2028 when we move from... So so look. Way... Wayve is working on these systems that are these l two plus driver systems where you can take your hands off the vehicle. You can get in vehicles today and experience this like in like in Teslas, and customers in general love them. And Wayve is working very hard to build credible evidence to show that that kind of driver can be safer than even expert humans. So that's really important. That's just the the foothold, though. What we need to do is move beyond l two plus to l three systems, where sometimes those are referred to as not just hands off, but eyes off systems. So those are systems where you can truly check out, do other tasks, and the system is responsible for safety. Now, of course, to get to those level of systems, you need to have much, much more strict stringent safety guarantees, higher bar because, know, can you imagine hopping in your car and saying, you know, take me to the South Bay and taking a nap? Yeah. That's a very different thing. It's, you know, almost at the level of what what we consider to be safe for robotaxis today. Wow. And I think when we get there, what happens for people is they get their most valuable commodity back, which is time.
00:18:17 · Ruby Justice Thelot, Brandon Basso
[00:18:17] S1: Their time. Their time. There is this really interesting stat where people's happiness levels are almost directly correlated to their commute time. Imagine, instead of being stuck in traffic from Sunnyvale to San Francisco for two hours a day, you spend that time working or napping or catching whatever you can do really exciting stuff. I know we're kind of at time. I wanted to ask right before we go, what is next for Wayve? What can people expect to see over the next few months? [00:18:47] S0: Get the system out there. [00:18:48] S1: The system out there. [00:18:49] S0: Economy for everywhere for everyone. It's to tell our story. It's to represent why we think that this is the right way to solve the problem. It's it's honestly to do posts like this and talks like this to sort of move the industry forward and be a thought leader in the space both with regard to the safety of the driver itself, but also how we plan on doing validation so that we can start having a conversation with the public and with our customers about how we plan on proving that these systems are gonna make our lives better.