I was in San Francisco a few weeks ago, and I did my first Waymo. You use your app—it’s the Waymo app. The car comes, you open the door, and there’s nobody in there. Then you start driving. I thought it was actually very comfortable because I didn’t have to discuss anything. I didn’t have to—I’m serious. Sometimes you have good conversations, but just as often, you’re not in the mood. I liked it a lot. Tesla is trying to do the same with their Full Self-Driving. And today, of course, it’s not full self-driving, but they’re getting there. In the latest versions, you don’t have to have your hands on the wheel anymore—you still need to look—but it can actually take you from your home to another place through the streets, avoiding things.
And you know, they’re behind their Elon time as they call it. But we’re clearly, I believe, looking at the Waymos. By the way, are you guys aware they’re doing 100,000 rides per week now? They started in LA—have you seen them driving here in Atlanta?
So, these are not fully self-driving yet. There are people in there, and that’s because of their mapping. They’re working on mapping and discovering all the oddities of how different it is to drive in Atlanta versus San Francisco.
And this is the world where Will is living. So, Will, introduce yourself and talk about your path to self-driving cars from where you started.
Will Bryan
Yeah, well, I’m Will. I don’t have the name tag—I guess I missed the table. But my first ride in a self-driving car, I think, was much scarier than Jean-Luc’s. I used to work in a research lab after school. I went directly into a research lab developing the algorithms for self-driving cars—for passenger vehicles, like the one he rode in, semi-trucks, and eventually autonomous racing and defense applications as well.
But yeah, I think my first ride in a self-driving car was when I was a master’s student. I had written some code to control the steering, throttle, and brake. We got in, and I was the test driver. It was actually going well for a few seconds, and then, the next thing you know, it started to veer off the road. I had to grab the steering wheel and save it.
So yeah, I spent years in the research lab developing the control and decision-making algorithms for self-driving vehicles. We were the first to do autonomous trucking on Canadian logging roads in the snow. Well, logging roads in general, but also in the snow—which is an added challenge.
And through all of this experience, especially that first one, what you learn is that you need to test and make sure these systems are safe to a certain level before you put a human in the car.
And the way that’s done in robotics and self-driving is through a virtual world—a simulation. Experts pretty much agree that somewhere between 5 and 11 billion miles need to be driven by an autonomous driver before it can be guaranteed to be safer than a human driver. And that’s obviously not feasible. Even at the government reimbursement rate, not including a test driver, it’s just not possible.
So, simulation and testing in a virtual world are really the primary tools used by those developing autonomous robotics. Through that experience, I eventually had the opportunity to get involved in autonomous racing. There was an international competition that brought together 40 of the best research universities from around the world in autonomous driving. I convinced the university to let me start a team—and we did.
Will Bryan
And this car here that you see driving around in this sporadic marketing video has no driver. If you look closely, there’s no helmet in there—it’s just a computer and sensors.
And with that, the theory has been that autonomous cars are always just one year away. Right? Elon says every year that next year, with Tesla Autonomy, you’ll be able to take a nap. And people, I think, take it seriously—then they actually do take a nap, and that’s when these crashes happen.
But the real problem isn’t normal driving—that’s been solved for five years. It’s all these edge cases—when a child runs out into the road, when someone runs a red light, or when the weather conditions are horrible.
And so the theory was that with autonomous racing, if we bring the best 40 research labs in the world and make them race with no driver, we can put them in edge conditions all the time and push that technology forward. They eventually narrowed it down to the top 10, and we were one of them. Then, we led that program and became the top-finishing U.S. team in that competition.
Well, you might think that with no human driver, the stakes would be lower—which, in some ways, they are. But these are very expensive, multimillion-dollar cars. And we quickly found out that even one small mistake can cost $100,000—you know, if you crunch up just one little corner of those cars.
So, simulation wasn’t just important for human safety, but also for financial safety.
And so, all of that very long, wordy introduction is to say—there was this huge problem. We had used every simulator on the market because we were a very well-funded lab, and they were all terrible. Everything that we tested in there would behave completely differently when we took it into the real car.
And that was the sole motivation to start Autonoma. Really, that was the only plan—it was not well thought out. It was just, let’s create something better because this had been such a pain point for us and everyone we knew.
So, yeah, then we—
Jean-Luc Vanhulst
The simulations are hard.
Will Bryan
Very hard, yes. Yeah.
Jean-Luc Vanhulst
What are we seeing up here?
Will Bryan
So, can anybody—does anybody know what this is? If you’re looking at it? It is LIDAR. 100 points to Gryffindor.
So, the thing about simulation—like I mentioned—is that it has to be an accurate representation of the real world for it to have any value. And that’s where we saw the biggest gap.
Now, we had spent all these years doing research and getting master’s and PhDs in vehicle dynamics, sensors, AI, and all these different things. We wanted to take that knowledge, which basically taught us that everything on the market was bad, and use it to create a commercial product.
Will Bryan
And so what we did is we took that research-level knowledge and fidelity and put it into a product that simulates the dynamics of a vehicle, the environment, and all of the sensors. This is one example: LIDAR, which is one of the most complex to simulate, and we basically created the most accurate virtual representation of the real world. So, right here, one of these is simulated and one of these is real. Can anyone tell which one’s which? You have a 50% chance—hopefully not. But as you can see, we’ve been able to replicate not just the visuals of an environment like a video game, but also the raw data that comes out. This is gigabytes of data every second that these vehicles generate. This is three LIDAR sensors being simulated here.
Will Bryan
But this vehicle in particular has seven cameras, four radar sensors, four GPS systems, a couple of IMUs, and lots of compute and networking that all has to be simulated. LIDAR is basically a bunch of lasers being shot out, and then it tells you exactly, in three dimensions, where everything is in the environment.
Jean-Luc Vanhulst
It’s a kind of radar, but much more detailed.
Will Bryan
Yeah, instead of using radio waves, it’s using light waves. So, not to get too much into the weeds, but we started, and were able—like I said, we had no plans—we just bootstrapped and were able to get our first customers who had the same problem as usual. We really started solving that same problem for others.
Jean-Luc Vanhulst
Yeah. And so, where does this make sense, and where are your clients now that you’ve gone into the market and beyond autonomous racing? Because there’s a lot of autonomy these days, and there is a lot of need for safety.
Will Bryan
Yeah. So, what we’ve found is that there are the OEMs and autonomous vehicle companies like Waymo that are out there doing autonomous driving, but there are actually a lot of other industries that, while maybe less flashy, are actually much more mature in their adoption of autonomous robots. One example here: airports, which are starting to test and use autonomous vehicles for baggage handling, detecting and retrieving foreign objects, and for security. The same thing goes for warehouses, manufacturing—they’ve been using that for years. Mining and all these other industries that people don’t think of when they think of an autonomous vehicle—these are actually much more mature. And luckily for us, none of our competitors are actually looking at these markets at all. They’re all focused on the automakers.
Will Bryan
And so, that’s where we’ve started to focus on a different area, bringing the same exact technology that we showed but applying it somewhere that can actually affect people, I think, earlier on and where it’s more ready.
Jean-Luc Vanhulst
Yeah. And how about, like, infrastructural? Like, you’re— I think you’re working very near here with a new client or almost a new client.
Will Bryan
Almost a new client. You know, the legal side can take a bit, but we talk about it here. Yeah, this is all confidential for the moment, but we had always been coming at it from the side of the vehicle, simulating the vehicle. But to do that, you also have to simulate the environment. What we found—and what has actually been really surprising—is the response we’ve gotten from the entire other side of the coin: the people who own the environments, the people who own the infrastructure. So it’s not just the car going through the city, but it’s also the city itself.
So, nearby, there’s a town called Peachtree City or Peachtree Corners, which is very advanced in integrating smart infrastructure and sensors into their city.
There’s been a lot of autonomous vehicle testing there, all of which has gone very poorly, at least in the beginning, because they didn’t have an accurate representation of their city for people to test in. So, these conversations started out with, We can make a digital version of your city. Then, all these people, like Waymo, and mobility companies that want to come in and deploy, can first test in that environment. That way, when it goes out and the leaves fall, they don’t get confused because that doesn’t happen where they’re from.
And where that’s actually led is that there are all these other applications of the exact same technology we’ve developed in other areas. So here’s an example: they’re looking to replace a really dangerous intersection, one where there have been a lot of high-speed accidents and even fatalities.
Will Bryan
And so, they’re looking to replace it with a traffic circle. What happens in the U.S. when you put in a traffic circle is that all the citizens get upset, and the city council hears about it. And so, what they want is to be able to show and demonstrate exactly what it’s going to look like and the actual impact it can have.
So, we’re partnering with them to help simulate the kinds of accidents and collisions that are happening currently, which they have all the data on, and then show what it will look like in the real world. Not just statistically, but with the actual physics and dynamics, being able to see what those potential dangers are. The goal is to demonstrate how massive reductions in fatalities can be achieved if they change it.
Will Bryan
And so, it’s really exciting because we’re talking about one intersection here, which is where we’re starting with them. They have an entire city. You can also extrapolate that out to basically every road. We’re also talking with the Georgia Department of Transportation. Their biggest challenge, their biggest issue, is accidents. They want to reduce accidents, obviously, to save human life, but also because every time an accident happens, they get sued.
So, if they can simulate different layouts during the design process, like maybe they need a barrier here, or maybe they need a different type of intersection, they can prevent those accidents. And when they do get sued, they can show that they did their due diligence and had the safest design out there.
Jean-Luc Vanhulst
Yeah. And by the way, does anybody know how many people die in traffic in the U.S. every year? Any estimate? Forty thousand. That’s the number. Thank God it’s not 300,000 or more, but 40,000. By the way, if you compare it with any European country, it’s at least double the numbers in other countries. So, it’s a huge number. And there’s a lot of, especially, I believe autonomy will probably get rid of, you know, 95% of those.
Will Bryan
They say 95% are human factors.
Jean-Luc Vanhulst
Yeah. Because people, you know, they might make mistakes autonomously, but they will never be drunk. They will never be distracted. They will never have a fight at home before getting in the car, or a fight in the car, or texting, or all of that.
Audience :
Does that mean we’re getting more rotaries around the country?
Will Bryan
Well, what we’re trying to do is make the vehicle safer and make the road safer. So we’re doing both sides of the coin. I think we probably will be seeing more and more traffic circles, because they very greatly reduce accidents. I mean, that’s not— we’re happy to simulate it. That’s not where we’re lobbying for anything. But, yeah, there’s been lots of towns. There’s a town in Indiana that replaced all their intersections with traffic circles, and now they have zero. Yeah, yeah. So we have, you know, we’ve built an entire traffic engine so all of the agents—it’s driven by AI. I know. That’s why we’re here. Everybody wants to talk about AI but all of the agents. So, we simulate human behavior. We simulate the pedestrians. We simulate, obviously, the autonomous vehicle.
Will Bryan
We have human interface. So, you could drive through Atlanta traffic. I mean, here at Peachtree Corners, they’re also putting in an apartment complex, shops, and a hotel that’s going to add traffic.
Jean-Luc Vanhulst
These are the two scenarios. That’s right.
Will Bryan
Yeah. We can simulate the entire city.
Jean-Luc Vanhulst
Yeah, you can.
Will Bryan
Exactly. And it’s not just statistical models that have been done in the past in 2D. It’s real 3D, and they’re reactive agents. And we can throw in—the other thing we’re doing that I know we haven’t touched on yet, that’s completely unique to us, is we’re fusing in the live traffic data that they already have from these intersections. So, we’re putting those agents into the simulation. So those are real. You know, talk about AI, that’s a bunch of human AI that’s in the loop in the simulation. And then we can throw in another 500 vehicles that aren’t actually there, and they can all react, and we can see how that behaves. So that’s something that we’re actually demoing in Texas next week.
Will Bryan
If anyone doesn’t have plans on, I guess, either Wednesday or Thursday of next week, we have a private event going on with one of our customers in Austin. It’s going to be super cool. But, yeah, it’s fusing the live data and the synthetic data to now be able to not just statistically try to determine what it’s going to be like in the future. We can show it.
Jean-Luc Vanhulst
So you could literally, basically, take the section with the traffic, with today’s traffic data in terms of volume, and then run the traffic through to the roundabout instead of the live traffic. And I guess one of the ways to sell roundabouts is probably the fact that traffic is also going to be just much more, less stuck. Right. Because that’s one of the biggest advantages too, apart from safety. It’s also just more fluid, right?
Will Bryan
Yeah, you get less of the contractions and stuff like that.
Audience:
So you’re what takes the billions of hours needed to go train something down to less. Is there a way to accelerate that learning curve?
Will Bryan
Yeah. So the reason that you need 11 billion miles of driving to test is not because you actually need 11 billion miles, but it takes 11 billion miles for you to get all of those random edge cases and variations. So now we can target exactly those things that are super rare, that maybe happen once in your life or never happen. Like a drunk driver running a red light as you’re going through at night, or a child running out on the road. All of those things. We can take those and test specifically for those and test them in every weather condition and every variation of that. And so now you don’t need 11 billion miles. You maybe need 100,000 scenarios. And we can now also, we’re cloud-native, so we can test that in parallel at scale.
Will Bryan
And what would take 100 years to actually test on the road can now probably be done in a week.
Jean-Luc Vanhulst
Yeah, you can simulate the ride from hell. Basically. That is only edge cases. Instead of having to drive so many.
Audience:
If every car magically had all the lidars on it and was pumping that data back, would that just accelerate finding all those scenarios so that you could simulate them as well?
Will Bryan
Exactly. You’re actually connecting the dots for me. So a city like Peachtree Corners and a lot of these cities that are trying to be more advanced and they’re putting in lidar and cameras and all their intersections, we can take all of that data and then automatically be generating those scenarios. Instead of an engineer or a group of engineers at, you know, Ford trying to think of everything that could happen, we actually now just have everything that is happening and then we can make all the variations off of that.
Audience:
Hey, so my camera and I work with a company called CRH. We’re the largest road builder in the US across.
Will Bryan
We should talk.
Audience:
So a big problem that we have is really we. We see that there’s a lot of innovation going on with design work and theoretical changes to roadways. However, when projects are designed, oftentimes they are designed years prior and they’re put on a shelf, and then they’re put out for tender bids, and companies like mine will go and bid on those projects and then build them. We know that there’s a better way to build them, but we can’t take them along for that journey because it’s already been pre-designed. So I’d be curious for your conversations in Austin and more generally Peachtree Corners and others, how can you take their design and say, guys, this is an old design. There’s a better way to build?
Jean-Luc Vanhulst
If you did it this way, maybe.
Audience:
Don’t change the budget too much. But I think that influence factor in there will be really important. So I would love to talk more about that.
Will Bryan
Yeah, for sure. Let’s talk after this. Yeah, we literally were having a very interesting conversation with GDOT two days ago about this specific thing. Why do they sit on the shelf?
Audience:
Just budget, basically. So they will pre-budget when I say they. The departments of transportation will be budgeting projects, and so they’ll design these projects for future need, and then when they’re ready for the fiscal year, they’ll be put up for tender for basically the lowest bid. And so that’s one other thing that the lowest bid is oftentimes the winner. The safest answer.
Jean-Luc Vanhulst
Right?
Will Bryan
Yeah, it’s the opposite of defense, where the highest bid always wins.
Jean-Luc Vanhulst
Two questions.
Audience:
Yeah, a slight detour on the question here. I mean, you must consume an enormous amount of computing resources as companies are building AI. Can you model out your costs? Is that possible these days?
Will Bryan
Yeah. So one of the things that’s, I guess, one of the benefits of using us is we’re extremely efficient with the simulation. So one company we’re working with is Booz Allen Hamilton. You may be familiar. They’re a large defense prime. They have an AI division that is using our simulation to develop, with pure reinforcement learning, autonomous drivers that are better than a human for all kinds of applications, as you can imagine. But to do that with those sorts of AI methods, they’re having to run 10 or 15 million simulations to develop that. And so what we’ve done is we found that all of the physics engines that were out there, all of the things that kind of underline this technology, were all inefficient and they couldn’t do the fidelity we wanted. So we just got really smart people and built our own from scratch.
Will Bryan
And it’s much more efficient. So on like a normal laptop like the one I have, you can run our simulation at 20 times real time. And so if you’re thinking about the cloud, you’re paying for the amount of time you’re spending. So if you can run 100 cloud instances at a time and then you run them at 20 times real time, then that’s, I mean, obviously a huge multiplier on the amount of simulations you can run for the same cost. Which one? Like maybe the biggest motivation for starting the company was one of our competitors. I won’t name them here, but they were running a simulation that was 20 minutes of real time, and to do all the same type of sensors with much worse fidelity, it took 16 hours to run what would have been 20 minutes in the real world.
Will Bryan
And that just really held everything up. And if you’re trying to test at scale, yeah, you’ll just run out of money. And so yeah, it can depend on the application. But that’s also where, you know, airports and ports and manufacturing plants that are closed environments. It’s one of the reasons why I think they’re able to be one of the earlier adopters and they’re able to have more benefit and more of a business case because in that closed environment, you can really constrain what you need to test and what the potential risks are.
Audience:
Question, challenge standpoint for you is that largely when you’re presenting, you’re talking to whatever it is, behavioral or you know, do you feel like there’s an organization that I’m never going to dip on the car? How much does that impact your selling process or do you have people over that?
Will Bryan
Yeah, I think the challenges in sales are really different based on the type of customer. If we’re talking to the people developing the vehicles, they know, I mean they’re already using something for simulation and they’re already usually very unhappy with it. And so we’re not really having to educate them there. It’s just trying to prove to them, when all these other companies they’ve tried have kind of lost their trust because it wasn’t as good as they said, is trying to prove to them that ours can do what we say. Now, dealing with the people on the infrastructure side, this is a new market for them because they haven’t been thinking in this digital world and they haven’t been thinking in this future-looking way. And so they’re blown away because they’ve never even comprehended it. But that takes some time for them to be educated and understand and allocate budget for it. So I’d say the challenge there is, I mean, there’s no competition and the value proposition ends up being really clear. But it’s just they were never even… It wasn’t even on their radar before.
Host :
We have time for a final point.
Jean-Luc Vanhulst
So I think it’s been mentioned before, but 1996 is a big year for Atlanta. 2026 is also going to be a huge year, I think, for Atlanta with the World Cup soccer coming. And that is also something that’s in… That’s where local city governments are interested in simulating.
Will Bryan
Yes. We’ve been talking with LA and New York both about the Olympics and the FIFA World Cup coming up on what they can do because they have so many of these problems that can be solved with simulation. It’s not just how do they route traffic, it’s where they put barriers. What are the sight lines for security? Where do they need to put these sensors? And so, yeah, when it comes to Atlanta, I think we’d love to be having the same conversations here about, for the World Cup, how can we help the city prepare and get the most value out of it and have the least congestion and the most safety around that?
Jean-Luc Vanhulst
And you’ll have a little bit more time to work on the simulation of the Dutch soccer field fans in the streets doing the wave. I don’t know if you’ve seen the…
Will Bryan
Orange wave and the flares. Yeah, I’m a big F1 fan. So, every time it’s the Dutch Grand Prix, there’s the orange smoke coming up.
Jean-Luc Vanhulst
Yeah. But it’s also, yeah. The pedestrians in the streets [are] doing the wave. So that’s a good simulation challenge. There you go. Hey, I think with that, we can close it.
Thanks for being a part of the community of courage by listening to the visionary founders and investors on the Atlanta Startup Podcast. Subscribe now so you don’t miss a single episode of the over 200 investors and founders sharing their insider tips and secrets to growth. Our regular listeners tell us we’re the briefing room for the innovation economy in the fastest-growing region of the country, the South –and when you subscribe, you become part of the inside circle.
The Atlanta Startup Podcast is proudly hosted by Valor VC. Valor is a venture capital firm that leads seed rounds in AI and B2B SaaS startups. If you like the podcast, check out more of Valor’s programs for courageous founders and investors, like Startup Runway.
Over $100M in early-stage venture capital and counting is catalyzed through Startup Runway’s grant-making program for pre-seed startups. Go to StartupRunway Dot ORG to learn more and apply directly for non-dilutive capital.
Valor celebrates VC DAY, the largest early-stage private capital conference in the region, at the end of the year. The top founders in the region, leading VCs, endowments, and family offices focusing on venture capital outperformance attend. Learn more at VC.Day.
At Valor, courage is the currency of innovation and the heartbeat of our culture. Thanks for listening and come back next week.
