Skip to main content

The 25-year-old billionaire building the future of self-driving cars

Luminar’s Austin Russell on the technology that will drive the future

Share this story

Photo Illustration by William Joel / The Verge

Austin Russell is the 25-year-old founder and CEO of Luminar, a startup in Silicon Valley that makes LIDAR sensors for self-driving cars. LIDAR technology had been used for short-distance mapping, but Luminar claims to have a functioning LIDAR that works at 250 meters, which is a breakthrough. Luminar recently went public, making Austin today’s youngest self-made billionaire.

And when it comes to self-driving cars, youth is definitely an advantage — Austin told me we’re still years if not decades away from fully self-driving cars, and there’s a lot of work to be done to make them safe, effective, and ubiquitous. That work is racing ahead — Luminar has deals with Volvo, Audi, Toyota, and others — but building a complete self-driving car is still a long-term project.

Here we go.

This transcript has been lightly edited for clarity.

Austin Russell, you’re the founder and CEO of Luminar. Welcome to the show.

Thank you. 

I’m very excited to talk to you. You are, as far as the last thing I read, the youngest self-made billionaire in America, your company just went public in a SPAC [special purpose acquisitions company]. How old are you?

I am 25 currently. And come Pi Day, 26.

You were born on Pi Day?

Yeah, exactly. March 14th.

It’s just a lot of things coming together for you there. That’s great. And auspicious. 

So, Luminar, it’s a company that makes LIDAR sensors. You have a number of deals to supply LIDAR sensors to major automakers. I want to talk about all of that. One thing that I always get frustrated by in origin stories is no one ever really talks about act two. In 2012 you were at Stanford, you had this idea to do LIDAR sensors. Something happened. Now you’re a public company with hundreds of employees and you’re a billionaire. I want to talk about act two for a little bit. Just that middle part of going from “I’ve got a great idea,” to “This company is actually up and running and functional.” So give me a sense of, at the beginning you were a student at Stanford, you got a Thiel Fellowship from Peter Thiel. What was the next step? Did you sit down and build a LIDAR sensor? Did you have the components for it? Were you like, “I’ve got to make a laser.” What was the next step?

I think when it comes down to it, there wasn’t any kind of miracle step, so to say, involved in that but rather these were already somewhat ongoing projects, even at the time when I went into Stanford, when I got the Thiel Fellowship, everything was already somewhat in motion. But it all starts with that innate passion and desire to be able to want to build, to create, to innovate and invent. And in my case I just had a specific interest around optics and photonics and other types of systems that can end up making a huge difference in these kinds of new products and saw the application opportunity for autonomous vehicles.

So [I] really just went all in on that. And it’s one of those things that if you knock it out of the park, it really is a trillion-dollar industry opportunity with — not to mention the humanitarian aspect of 1.3 million lives lost on the road every year, to be able to have an opportunity to impact that and make a difference. It maybe seems like a crazy long shot at the time for that to happen and materialize. But at the same time, I think you just have extreme focus around what you want to do, and continue to solve for it, build for it, learn from all the folks around you and just continue to deliver. 

This is the thing, this is the act two. You got the fellowship, you got the check, what did you do the next day?

What did I do the next day? That’s what I’m saying, it’s like the same thing that it was otherwise doing. I don’t know how much it fundamentally changes other than accelerates the exact same thing you’re doing. And you just keep accelerating, and accelerating, and accelerating, and you have this compounding exponential effect that you can grow and snowball into something much bigger. And that’s really what happens. And by the way, for things like the Thiel Fellowship, don’t get me wrong. The 100K check is great, but there’s multiple ways you can generate 100K income to be able to start a business, the real value is the network. How do you learn from the most successful entrepreneurs on the planet to be able to build this kind of business and scale this kind of business? And like I said, it takes a lot more than a great technology and concept to build a great business and that’s what’s required if you want to actually have your tech see the light of day.

It was just you at the beginning, you had this interest in optics and photonics. You saw the market opportunity for LIDAR, particularly in self-driving cars. At the time, 2012-ish, those systems were really big. I think Luminar’s advancement is that you’ve made them smaller and easier to manufacture. You’ve made them more of a useful device. Here’s a simple question. Who was the first person you hired?

Oh, a couple of key folks. One of them, I worked with originally from a consulting perspective, and then ultimately brought on as a co-founder into this. So it’s a guy named Jason Eichenholz, who’s our CTO, and responsible for a lot of now forward-looking R&D programs and projects that we have going on from the LIDAR standpoint. He actually helped to establish the Orlando, Florida office that we have outside of the Silicon Valley presence that we had. And that was where we really started to scale up the hardware team out there and were able to build out a capability. It turned out the highest concentration of these types of LIDAR engineers out of anywhere in the world was in Orlando, Florida.

So, you go where the people are. It’s a Space Coast, defense industry, Cape Canaveral and all that, that I really brought in a lot of the talent from. Actually another interesting one, I did end up hiring my uncle, who was an electrical engineer and someone that I had admired. I didn’t know him super well previously, but my parents pointed it out and said, “Hey, you know your uncle does that electrical engineering thing.” And I was like, “Yeah!” So he’s certainly good at rapid prototyping and definitely helped out early on as well.

So just walk me through: you’ve got your early prototypes, a small team of people, you think it can be productized. How’d you go from that to, you’ve got a manufacturing operation at scale that can obviously support the size of a public company now.

Yeah, so it definitely is a shift going from the core technology development, which we did for the first handful of years, to being able to have a scalable operation that can build automotive-grade product into series production. And the most important thing that’s always been our goal and focus, the reality is that even, effectively, all the autonomous vehicle companies today that are built building products, it’s really much more meant for R&D than it is for something that’s ever actually going to make its way into a production vehicle.

And that’s a key distinction here, of where you have to follow automotive process, and the work around how you can build these sensing systems up. And that’s where for us, we leverage the core expertise around how you can successfully assemble these and doing the process engineering and the manufacturing engineering and everything, to be able to get to a stage where we can have a formula for not just the technology itself, but how to be able to build the technology and build it at scale. And that was where we also built out that capability, in Orlando.

We have about 100,000 square feet worth of space that we’ve been using for our facility. That’s where the manufacturing process that we have is developed, which ultimately, we will transfer and are transferring to the relevant contract manufacturers to be able to do the final assembly for series production. So that way we’re not taking on thousands upon thousands of employees ourselves; as we scale up there too, we get to leverage outside resources for the more “commodity” aspect of assembly of the device itself.

Whereas the formula, the automation and how you build it successfully, that’s important to what we’ve had to do to keep in-house. So it’s a hybrid model. And that’s something that just takes a lot of work, a lot of expertise with the right people to build it up. And that’s always been core to the philosophy by the way, is make sure you get the right people on board that have huge domain expertise in the key areas that you really need it in. It’s easier said than done, but at the same time, if you have the right people there, it can scale that like we did in this case, it makes all the difference.

How much time do you spend with your engineering team doing engineering stuff versus with the business and commercial side of the company?

Still quite a decent amount. Obviously in the beginning it was all that, 100 percent just diving deep with the team. As we scale up, it’s still absolutely critical to not just keep a pulse on product and product development, but still to be able to shape that with the same vision that you have and make sure that stays on track, make sure that we can easily drive forward with all the new inputs that you have and continuously shape the future of how you see that playing out. And that’s something that I absolutely still am hands on in. I don’t think that’ll ever change. But there’s no question that there’s a lot of balls to juggle. There’s a lot of stuff that has to happen, but they’re all synergistic. You can’t do engineering in a vacuum, or product management in a vacuum, or commercial work in a vacuum. They all have to tie closely together. And this is where it’s critical to have the frame of mind that combines all the different areas of the business and all the different aspects of a product and everything that comes into play and having the technical understanding of that as well, which is not always a given, in order to be able to succeed and scale.

As you’ve been growing at the company, one of the tropes of early internet company founders is that they hired grownups. Mark Zuckerberg hired Sheryl Sandberg, Larry Page and Sergey Brin hired Eric Schmidt. It’s just a cliche. Have you put yourself in that mode? Have you thought, I need to hire some grownups that are going to help me go public, I need to hire some grownups that are going to help me operate a large company, or have you been more targeted?

You know, it’s a great question for that. The key thing here is that it’s something that we’ve been doing. It’s something that I’ve been doing, to be frank, this whole time throughout our journey. As it relates to step-by-step through all of this, I don’t think there’s any single person that is actually hired to take the reins of the grownup doing X, Y or Z, rather [it’s]: Build the rock star team of all the respective leaders in their field, and build a cross-functional org structure where you have a clear set of responsibilities for each one of these folks. Where it’s really my job to synthesize all of that information and be able to make corresponding business and technical decisions based on that path, and rely on the team from an execution perspective, to be able to see that vision through. So it’s a little bit of a different setup than maybe you would have with some of those folks, but at the same time, not uncommon at all. You mentioned [Jeff] Bezos here, [he] just finally handed that off, which is great to be able to see. But [he] had a similar type of cross-functional org structure for a huge period of time throughout Amazon’s growth trajectory and curve, and among many other companies that really set that up.

You’ve got to get the right insight into the right folks in terms of how to set this stuff up. The hard part about this business though, is this isn’t like your average internet company here. Where they’re having people that have built internet-related companies and social media-related work and had executed on paths and visions for those different types of businesses. For this, this really is something that from a technological perspective, as well as business perspective, it’s never been done. Even just carving out the fundamentals of the business ball. When you take a look at, where does all of Google’s and Facebook’s revenue come from? Ads, really that’s what... You take a look at all these fancy, crazy technology developments and everything, and like, “Okay, what’s with this model? It’s ads?” This is a completely different level of complexity that requires a different level of cross-functional understanding, all the way from the fundamental aspects of how the technology works, all the way up through the staff, and that’s why it makes it a little bit different and very important for leadership and whoever is leading the company to be able to have that full staff understanding and be able to drive the business forward.

And I do think that’s why there have been some great celebrated leaders within the overall autonomous vehicle space. It’s not just purely based on that basic understanding rather than sometimes even any execution at all, but it’s just so valuable in this space to be able to have that understanding, it drives forward a lot of the different ideas in the industry. But that’s what I’d say is the case for me. I’m continuing to bring on other great team members, expanding the overall executive leadership team that we have, but it’s been a very effective structure for how we’re executing.

One question I ask every CEO who comes on the show is, what’s your decision-making framework? And I think that question is particularly interesting with you because you started the company when you were very young. You’ve obviously scaled to this point. I imagine you’ve learned a lot of lessons along the way; particularly, you’ve almost certainly hired people who are older and have more experience than you. How has your decision-making framework evolved?

I would say just over time, as you build and scale a business, you have to go from being comfortable with decisions with nearly 100 percent of the information to being comfortable with decisions with effectively, what evolves to a smaller and smaller fraction of that for given things. You have to have greater levels of delegation, you have to have greater levels of talent and quality of talent as you build it up, and you have to continuously just keep raising the bar in terms of what you’re going to be doing. And I would say from a decision-making framework standpoint, it goes at the same time in a parallel thread beyond just increased delegation, more shifted towards going from technology-driven decisions [to] just trying to do whatever you can to create the best product. It’s like, “Okay, well, what do you do?” You have the best product, how do you make sure it sees widespread adoption in the industry?

And that’s where it becomes focused around being able to successfully enable that over the long run and being able to establish strategic initiatives and plans to be able to drive that forward to realization. So that’s where you have to apply, always, a business or financial mindset to all of these things to make it work. Because if you miss even part of one of the pieces of the puzzle, you effectively end up with zero. It’s like a binary equation here. You have to get everything almost perfectly right. And that’s why this is just so impossibly hard. Going from the R&D stage to actually realizing this through into a real product and business case and everything around it is a huge challenge, so that’s what you have to keep your eye on.

You were talking about prioritizing the product and we were talking about decision-making frameworks. So give me an example, you were saying you’re shifting from building the product, building the prototype to strategic decisions to make sure you can get to market. Give me an example of a trade-off you’ve made along the way.

Well, the ultimate goal of the balancing act is to try and minimize one-to-one trade-offs and to be able to scale the business side as much as you can scale the technology side in parallel. But nevertheless I would say, for example, we have no shortage of ambition in terms of what we’re trying to solve or what we knew we could build and make a difference on, beyond just the LIDAR sensing system. And I think there’s probably some desire to accelerate our software roadmap beyond our hardware roadmap, even faster than we ultimately did. Actually, much faster. But at the same time, you have to be careful about stacking too many parallel programs together and too much parallel risk and taking it step-by-step.

And as exciting as it can be to have something that you know you can solve for, ultimately, that was a trade-off that we had to say, “Okay, remain focused, solve the LIDAR problem, see it through. Yes, we’ll continue to build up the software stack in parallel for ultimately a more turnkey solution to the autonomous vehicle problem, but take it step by step.”

And that’s effectively what we did, and it was a decently long journey, eight years with this. So I think that’s where the last few years went, we spent more time really ramping up the commercial side and ramping up the software side as well. One other interesting trade-off that we made — taking it from a totally different angle, from a marketing perspective a lot of times — I know being in stealth mode for some period of time is really cool these days. But at this point back when I first founded the company, a lot of times you would have people that are announcing every stage. And I think it’s still usually the case now, every phase of their journey, oftentimes before you actually have anything to show for it.

At some stage, it goes back to a core philosophy versus like show, not tell. Because it’s really easy to have a lot of claims. And by the way, there is no shortage of claims in the overall autonomous vehicle space. In part, it’s because it’s easy to say things, because you don’t actually always show it or have to show it. People have ways to easily evaluate if it’s actually real or not. It’s some of the reason why you’ve seen some of the recent activity from companies that maybe couldn’t even raise private capital, now getting to a stage where [they’re] trying to enter the public markets.

But nevertheless for us it was always heads down, focus on delivering, don’t let any noise affect you. And by the way, don’t try and give any people a reason for a head start here either. Oftentimes it doesn’t actually benefit you. At least initially, consumer or even OEM [original equipment manufacturer] awareness is less relevant because you have to actually finish the product first. You want the first impression to be really good. And that’s basically what we spent the first five years doing. Deep down in stealth mode, building out the core product and technology and getting to a stage where we could actually launch the company and scale this up. And that’s really what we did starting in 2017, five years in, after we had developed the laser, the receiver, the scanning mechanism, processing electronics, all of our own chips that we had in the system. And it came together, it all worked. Somewhat miraculously, somewhat unexpectedly, at the same time.

So that’s where I think we were able to capture the attention of some of the world’s largest automakers in terms of showing the art of what is possible. We had a system that had specifications that people thought there was no way this was even doable from a physics perspective, much less something that could be economically produced. So that was the start of where we gained a lot of interest from beyond just the technology, but from a commercial angle as well, and a key catalyst for scaling this up.

So I want to dive in on the first trade-off you mentioned, which is going too far, too fast. You said you held yourself back from overthinking building the turnkey autonomous driving system, where, I’m Patel Automakers, which is a longstanding dream of mine—

There you go.

And I’m like, “Man, I really want this car to drive itself. I’m going to put out bids to Luminar, to Velodyne, to Siemens, for a self-driving car stack, I’m going to buy it, I’m going to integrate it into my car and I’m off and running.” And I didn’t actually do any of that work myself. You want to be that kind of player in the end.

Yeah, absolutely. We want to be a turnkey solution for this, and to some extent have already started evolving into that for these different OEMs. And really the reason why is, to be able to build an autonomous vehicle, the LIDAR is just one part of that holistic solution. Now it is a key bottleneck that’s been preventing the deployment of, in large part, this industry just by way of having something that meets the performance requirements, that meets the economic requirements, that can actually be scalable here. But at the same time the LIDAR itself doesn’t drive your car. You need the software to go along with it, as well as some of the rest of the ecosystem there to be able to push forward with that.

Now there’s some argument that some would say of like, “Okay, what on earth? Why are you guys screwing around with software?” There’s like “Yeah, okay. You carved out your value proposition in hardware, but you have folks like Waymo and Cruise and all these multibillion-dollar software development companies that are taking that on.” And here’s the key distinction though. I think if that was applicable to all the same things that you’re doing, all the same OEMs that we’re working with? Yeah, we wouldn’t be in that game.

But the key distinction here is that seeing the difference of, for example, it was very clear that everyone, virtually every company in the autonomous vehicle space, has been focused on robotaxis specifically, and all the big guys there. And that’s their primary application, that’s what they’re developing the software for, but it’s actually very different developing the software for a highly complex urban robotaxi scenario ... and building R&D cars for that, because that’s the most advanced stage that they’re at today, with like $100,000 roof racks full of sensors and stuff, versus building something for a production vehicle in a more constrained application. Something that could be deployed in the relative near term, with an auto grade quality code that you can put your life in the hands of.

So it’s a different trajectory, and that’s why we really focused around developing this full-stack solution for OEMs, for the automakers themselves, as opposed to us trying to do some rogue deployment with a robotaxi company or like a ride-sharing company, like building our own network of vehicles. Don’t get me wrong, that’s going to be a great route. It’s going to be huge value at the end of the day that’s created based on that and based on ride-sharing, but we don’t see that feasibly happening even with ours... We have customers in that domain and we’re helping accelerate, but it’s still like a decade-long problem, [more] than it is something that’s going to be deployed next year.

Isn’t that just a function of cost? Or the cost versus the revenue opportunity? So if you’re Waymo, you buy a Chrysler Pacifica for $50,000, you put $200,000 worth of stuff on the roof to make it drive itself — No one’s buying a quarter-million dollar Chrysler Pacifica. I can’t imagine who would buy that. There’s probably some hardcore Pacifica blog that’s going to write this up now, I’m hoping — But you use that as a taxi. Your quarter-million dollar minivan starts generating money. You’ve built a business that eventually returns profit as you scale down the cost of the roof rack. You’re saying you’re going to take on all of that cost and then sell a turnkey solution to automakers so they can ship a $55,000 SUV to an average family that can drive itself.

Absolutely. The whole point is to have a solution. You take that $100,000 roof rack full of stuff, or $200,000 roof rack, and effectively put it into a package that’s more on the order of $1,000 than it is $200,000. 

So that’s a huge part of the equation. No question. Is the cost and the economics there? But it’s also just as much building a solution that’s tailored for production vehicles that actually can be put into a car in an auto grade solution, in capacity, and having the software that’s focused around the use case in the application and the domain. You’re going to have a completely different level of software complexity trying to build for an urban environment than you are for the highway environments that we’re focused on for the initial deployments with OEMs. And the whole point is, how do you do all of that on the sensing side? And the software side, you solve that problem. You also need to make sure that doesn’t require a supercomputer in your trunk to actually process.

And that’s where being able to get all of that work on the same kind of Nvidia GPU that people are embedding into their production vehicles, that’s key. And that’s effectively what we’ve had to solve for as part of the full stack solution. Now, is it going to be driving you everywhere from point A to point B, anywhere you want? Absolutely not. But is it going to be enabling some level of autonomy that’s reachable by consumers and dramatically improve safety on your vehicle that can be put out onto billions of production cars? Absolutely. So then that’s how you start generating a real business.

One of the things I always think about with what I would call foundational technologies: sensors, display technologies, batteries, is we tend to build software and products around the limitations of those technologies. And then sometime along the way, there’s a huge shift in foundational technologies and our entire belief about what the products can do changes. And we have to build entirely new software. A really dumb example is the transition from CRTs to LCDs. There was a whole conception of what the world of technology would look like when CRTs were the display technology, we shifted to LCDs and now literally everything is an Android computer. Do you worry that LIDAR is one of those foundational technologies that will get disrupted in that way? Or is it that end-all be-all of sensing around you?

It’s a good question. And I think it’s foundational technology itself, but also the economics of those technologies, the performance of those technologies, what are the theoretical limitations in terms of what you can actually achieve with each of these different systems. And I would actually say that LIDAR is the perfect example of what we’re just talking about here, of a technology that’s entirely disrupted the mindset in a business model around what was even possible from an autonomous perspective. For example, historically LIDAR has been very limited in terms of the range performance of what you can actually do, how far out you can see. That was part of the reason why people were focused on urban environments for their autonomous vehicle setups for the robotaxis operating at low speed.

Now, when you can see very clearly out to 250 meters into the distance, different story. You can start to operate on highway speeds. You can now enable a different operating domain and something that actually ends up being more simple if you have the range to be able to do it. And at the same time, for example, the economics: If you have something that isn’t that huge $100,000 roof rack full of stuff, and you can actually have it embedded in a production vehicle for something more on the order of $1,000, then it makes all the difference.

We’re pricing our sensors today at that mark for many of these applications. So that’s where, seeing this actually happen in a production vehicle, it completely shifts the mindset in terms of what’s possible. And I’m not even sure that all of these companies would have the same business model that they do today if they knew this was possible from the beginning. I actually don’t think that a lot of the robotaxi companies would at all. But when it comes down to it, anything beyond that, if you’re talking existential perspective, about not just LIDAR disrupting LIDAR, or something disrupting not just improvements of the technology, but disrupting it altogether, the reality is that the autonomous vehicle problem, and being able to try and solve all of these different edge cases, is really, really hard.

People try to say, “Oh yeah, we’ll solve it. It’s easy. We’ll figure it out in 2022 or ‘23, whatever it is.” No, this stuff is really hard and no one has ever actually solved it at this point to date for the kind of application domain that you’re looking at. Now, you really need to throw everything you can get out at it. There was definitely an idea a few years ago that it would be something that would be solvable by just time and data collection and just what it takes. But the reality is it just requires a lot of raw, intense performance of a system to be able to understand what’s going on. And required even much better performance from a LIDAR standpoint, to be able to enable the capabilities, to have your understanding of everything going on around you very far out, and with the level of resolution performance that’s needed.

On top of that, of course, there’s continuing software capabilities that need to improve. It is improving, it’s getting better, but the software is not going to be there for a while, for more of these urban environments and constraints, regardless of what anyone might say, if you want to try to have a car go from point A to point B in these environments. It’s going to take time even with our sensors, it’s going to take a lot of time. But the point is how do you get this out there sooner than later? That’s where you’ll go for the more constrained environments.

Now listen, of course when talking about disruptiveness, you sometimes have confusion that can arise— there’s a very bright line between there. There’s assisted driving and there’s autonomous driving. One requires the driver constantly in the loop, ready to take over the wheel at any given moment whenever it makes a mistake, as it’s following a couple of lanes on the road ahead. This is the equivalent of things like Tesla autopilot — or the systems that the company Mobileye has been providing to OEMs for the better part of a decade now. They don’t always enable the same full functionality that Tesla does more aggressively — versus true self-driving, true autonomy. Hands-off, eyes off, read a book, use your phone, work on your laptop, watch a movie, take a nap, that kind of thing.

That’s autonomy and that’s the key step function value. And that’s what this enables, that’s the whole point of what this LIDAR enables. You don’t need a LIDAR at all if you want your car to follow a couple lanes in the road ahead of you and keep your hands on the wheel and eyes on the road. This is what it enables and it doesn’t matter what you call your assisted driving system. You can call it autopilot, full self-driving, whatever it is. Or like I said, the other OEMs that conservatively name their systems more appropriately.

So, that’s the key distinction there. The more performance you get, the safer the vehicle is. And I don’t think anybody’s going to stop and say, “Yeah, I think we can just decrease the safety of these vehicles by 100X.” Even if you got to a point where you could achieve that human-level safety threshold, why would you make your vehicles that much less safe? It doesn’t make any sense. So that would be not just in the near term, but I’m talking even a 50-year type time horizon. You don’t really see it being disruptive. You see, how do you continue to strive for even better performance?

Right. You’re saying if we made a fully autonomous car now that drove as well as me, a human, that’s not worth it. Because on average, humans are not good drivers.

Yeah. So you’re saying with LIDAR or you’re saying without LIDAR, for example, just in general.

Just in general. As a goal, as a marker. We made a car, it doesn’t have a steering wheel, it can drive as well as the average person, probably.

Absolutely. No, 100 percent. Now, you have to start somewhere with it. But the point is, is that people aren’t going to stop there, it’s not going to be good enough. Good enough, at least in our view, is driving the 1.3 million lives lost every year. By the way, just putting that in perspective. That’s like in the same vicinity of the crazy pandemic that we’ve had over the past year, in terms of lives lost, and it happens every single year and it completely doesn’t discriminate against underlying conditions, or whatever. It’s insane how accepted this is, just in everyday life. And that’s where if we can have the opportunity to be able to solve for that and drive that towards zero, that’s the mission that we should be going towards, that we all need to be rallying around, as opposed to just the human level, that safety is rational. Now that’s not to say that, like I said, you’re going to be 10X better, or 100X better off the bat, but you’ve got to keep pushing for that.

So I cleverly opened the door to the Tesla conversation and you walked your way through it. So the reason I asked about foundational technologies, and I think the reason you brought up full self-driving or autopilot or whatever Tesla is going to call it, is Elon Musk, who I think his office is just out your back windows over there.

I can see it right over across the way.

He has famously said LIDAR is a local maximum, it’s the wrong approach. It’s going to hit some ceiling of performance and that’ll be that. And the right answer for vision in automobiles is having a bunch of cameras and doing it in software at a high rate. Next to that is the other thing you brought up, which is Google and Waymo saying, “We’re going to use a bunch of technologies, we’ll have an enormous amount of data that we’ll measure by miles driven, and we’ll collect all that data and we’ll run it through machine learning. And that is the correct approach.” And they’re very proud of that approach. But let’s start with Tesla, Elon Musk and the local maximum comment. Is his approach to just using a bunch of cameras, does that actually have a higher ceiling, or do you think that you can catch up to it?

So just a couple of things just to level-set with that. You gave the Waymo example here and the Tesla example. It’s all about, what problem are you trying to solve? Like, what application are you trying to enable? What feature set are you trying to enable? For example, from what Waymo is trying to enable, being able to build these robotaxis for urban use cases and solving for that, they’re absolutely doing the right thing. They’re doing everything step-by-step that you should do and executing on it, probably someone in the industry would argue better than anyone else. Which I’d agree with, or maybe at a quicker pace as it relates to that. Obviously you have to be conscious around what is the application and also what are the different subcomponents that are there, that it’s still a long road ahead.

There’s other things like higher performance LIDAR that help accelerate you there much more quickly and are something that you ultimately need to enable. But nevertheless, I think that’s the point for [an] urban robotaxi development program. When you take a look at something like Tesla, it’s again, how does what you build align with the product you’re trying to develop? And it’s the same thing about assisted driving systems versus autonomous driving systems. If you want to build a great assisted driving system, you should be doing exactly what they’re doing. Everything spot on, collect lots of data, build the model, build everything.

Now arguably, their system was originally powered by Mobileye some time ago, as with pretty much all the other major automakers, they broke out from that. It was initially worse. Now it caught up to what Mobileye had. But this is nothing like, fundamentally, a breakthrough, or novel in the context of having an assisted driving system that does these things. The real challenge is when you try and cross a threshold and call things autonomous. 

But the reality is that the threshold for an assisted driving system versus an autonomous system, it’s orders of magnitude different. It’s not even kind of close. It really is a factor of... If you want to improve to a point of where it’s safer than a human level, where you would never need to take over, you need to take that disengagement rate from once every 20 miles to once every 2 million miles plus — disengagement meaning you have to take over the wheel to prevent otherwise some incident or accident like in the current assisted driving systems.

So the whole point of where LIDAR comes into play is to solve that problem. Obviously anyone that tries to, or even just a single person that tries to promise that without irony, the hard part is, you back yourself into a corner if you’ve been saying the same thing for five years, and it will come out. It’s supposed to come out every year, the next year. But, I don’t want to have any respect loss for the incredible things that new players in the EV world and even folks like Tesla built and disrupted on the EV side, it’s absolutely fantastic. But at the same time when it comes to autonomy, we just need to make sure we get the terminology and what’s actually possible straight, because that’s ultimately beneficial to consumers. The last thing you want is people misusing the technology and thinking it’s capable of certain things when it’s not. So it’s a really hard problem. It’s a long way to go. Even with the best performing LIDAR, it’s still a long way to go, much less trying to make it a hundred times harder without it.

You think it’s 100 times harder without LIDAR?

That’s probably an underestimate. Now, okay, let’s clarify here. If you have a really low-performance LIDAR that only gives you like a handful of points out there in the environment, maybe it’s only incrementally more helpful. The whole point of the LIDAR is that it gives you a 3D map of everything around the world, in real time, you know exactly where every object is. The hard part with cameras alone is that they give you a 2D view of what’s going on. You never really know where the objects are. You effectively have to guess, and people have made software that gets better and better and better at guessing. And this is what Mobileye and what Tesla and these companies have been doing.

But the thing is that it gets to a point of where you approach like 99 percent accuracy on it. Here’s the problem, even at 99 percent, they’re still not quite there. But even if you get there, it’s still not even remotely close to being good enough. You need like 10 nines worth of reliability in there. And it gets exponentially harder with each nine. It’s like, is it acceptable to run over one out of every 100 people because you didn’t see them and guess where it is? Any rational person would say no. And that would not make it anything close to the same safety threshold as a human. So that’s the whole point around it.

And the LIDAR, assuming you have really high performance, long range, high resolution, that you have to have to get a clear understanding of everything going on, but if you know exactly where everything is, like we do down to a centimeter level of precision, it makes all the difference. And there’s no question it’s like, “Yes, it’s another sensor on board the vehicles, but there’s a good reason why effectively every autonomous vehicle company has been going down this.” And to your question on data, there’s no question that more data and better data absolutely helps, but the whole point is when you have a more limited quality of data set, for example, with cameras, it absolutely tapers off in terms of the utility of it. Like even OEMs, the Mobileyes of this world, only use a really small fraction of that overall data that’s collected.

And it has a marginal utility curve that only helps you so much. But let’s say, even if you were able to continue an amazing growth rate, like 50 percent improvement year over year for these kinds of systems, when you’re trying to solve a gap of like 10,000X in safety … it’d be like a century or two by the time you got there. So that’s the distinction. And that’s where I think there can often be a gap between the understanding of where this kind of... And I just mean more broadly, where autonomous-related technology is today and where it actually needs to get to, and how we successfully close that gap, because there is a huge gap.

Let’s put into context what you mean by LIDAR performance. So right now I think the easiest way for an average person to get a LIDAR sensor in their life is to go out and buy an iPhone. The Pro models of the iPhone have a LIDAR sensor on the back. They’re very small, but—

It’s so awesome, by the way.

It’s cool that they’re there. They don’t do a whole lot. They help with some picture stuff, and there’s a bunch of AR gimmicks, but that’s how you can get a LIDAR sensor in your life. Just put the performance of the iPhone LIDAR sensor on a spectrum with what you’re hoping to achieve or what you’re building.

Okay. So still, for example, an iPhone LIDAR sensor, you’re talking about seeing single-digit meters out into the distance. And of course in this case, it actually depends on whether it’s daylight outside, or whether it’s night. We have to see regardless of the condition, 250 meters out in the distance. So even if the sun is shining into the sensor. So you’re talking up to 100 times the range here of what you would see on an iPhone LIDAR, with orders of magnitude more resolution performance and something that’s in an auto grade mass producible device. By the way, it’s a square function. So the farther you go out, if you’re 100 times the distance, it actually requires exponentially more output power from a radiometry perspective of, you actually have to square that. So you’re talking like more on the order of 10,000 times the amount of performance that you have to have out of LIDAR just to be able to get that distance.

So that gives you a spectrum of comparison of why this is actually really, really hard. And by the way, you have to do all of this and see not just bright white reflective objects, but you have to see the really dark objects too. Because what happens if there’s a tire on the road, a black car, or a person in dark clothing, anything, those are actually really hard to see, to get light reflected off of, which makes it another 100 times harder too, than a bright white reflective object.

So these compound and compound. The whole point is there’s no off the shelf parts or components that you can buy for a LIDAR that enable you to do this today. It doesn’t exist. This is the whole point of why we had to actually build it entirely from the ground up, we had to make all of our own components, we had to solve for all of that in order to actually make this work. You have to do that again at the same time, while having to be economical enough that it’s not tens or hundreds of thousands of dollars to actually build, and to be able to put it into a production vehicle. So that was what we had to solve for. We were fortunate enough that we were able to, over that five-year period of really deep, intense work and meeting all those different, very stringent requirements.

And yeah, I think that having that be commercialized into production vehicles, that’s the whole thing of what’s next, is that once you meet all those specs and once you get the commercial step right, it’s all about getting into serious production. And we were fortunate enough to be the first and only company to really be able to enable autonomy in the series production from these different programs out there, starting with Volvo and then cascading to some of the other OEMs that we’re working with as well. So it’ll be a good start. That gives you an idea around the performance of what it takes to make it commercially viable as well.

I ask every self-driving car CEO this, and you don’t make cars, but I feel like you have a good handle on it. How long until I can buy a car without a steering wheel?

Probably I would say early 2030s, realistically. A couple of different observations here. ... There’s a key distinction to that question of, when can you buy a car without a steering wheel? It’s not clear that people will actually be selling these cars without steering wheels. If you get to a point where you don’t have to have a steering wheel, then it likely will be deployed in a more robotaxi-type application initially as it scales up and then ultimately be offered to consumers as well.

But here’s the thing, is that interestingly enough from a regulatory perspective, it actually gets a lot more complicated when you start ripping out steering wheels, changing all the different functionalities of the core of the vehicles to be able to establish what you could see for a futuristic version of autonomy on these cars versus just enabling autonomous capabilities on existing production vehicles. And that’s going to take some time. I would say, in theory you could probably buy a car without a steering wheel, that’d take you from point A to point B in a more limited geographic capacity, by between 2025 and 2030. But the question is, would you really want that, would you buy a car that can only take you to certain limited destinations, and it’s safe?

I feel like there’s a lot of people who would move farther away from New York or San Francisco if they had a car that could reliably and quickly move them back and forth from work.

Absolutely. But here’s the thing, that’s the whole point of exactly what we’re doing for 2022, not even 2030, 2022 on these different production vehicle platforms, is enabling highway autonomy. The whole point is, you take your car, let’s say you’re living out farther into the suburbs there and commuting into the city, take your car, manually drive over to the freeway. Once you’re on it, engage autonomous mode, hands-off, eyes-off, and be able to recover that time ...everything from using your phone to taking a nap, and then a few minutes before the final exit, have a planned manual takeover and then drive to the final destination. And that’s that hybrid of a problem and a solution there. Whereas if you eliminate the steering wheel altogether, then it’s only going to be able to work on whatever routes that it’s effectively pre-programmed to be able to work on. And that’s going to be a much tougher distinction there as it relates to that. So this is just going to be a normal car otherwise, that you can also use any day, anytime. 

It also has dramatically improved safety because it will now actively take over in an instance where it’s predicting an accident could occur as well, which is something that really hasn’t been seen before on vehicles. And another really interesting aspect of how you can start saving all these lives without having autonomy. You don’t have to have autonomy everywhere to start doing that.

You’ve got a handful of big partners, Audi, Volvo — which of those companies should we see Luminar technology in first, in a shipping product?

You know, Volvo is certainly there first with these programs. We’re driving towards basically a start of production with the programs sort of around the corner next year in 2022; by the end of that we should be able to deliver this autonomous solution that can be enabled, or really, I should say, these LIDAR-enabled vehicles here. Those are things that you can actually buy, and that is a key distinction when it comes down to it. So it’s something that is definitely going to be the first to market in terms of any level of autonomous capabilities that are functional and tangible as well, for that matter, by consumers or otherwise. So it’ll be exciting to help enable that to happen. We have a host of other great partners.

Even most recently, just in the second half of this last year, over the past handful of months, [we] announced Daimler trucks as a lead partner on the truck side, just as Volvo is a lead partner on the car side. And then also Mobileye is a lead ecosystem partner, including their foray into robotaxis as well, which I think is going to be super promising. So that’s where we’re delivering against, but Volvo really helps set the clock for everybody. And frankly, a lot of the rest of the industry timing-wise, from our product perspective and how we scale that into the rest of the other OEMs. But I think they’ve had a really strong safety and brand focus around this and just see this as too critical to be able to mess around with, and be able to have the best technology and the best program and make it economical for their vehicles. So it’s going to continue to get better and better over time, via over the air software updates. But it’s certainly going to be a great start to the whole new era of autonomy in the industry.

So we’ve had this long conversation about how you started the company very early, very young. You were in stealth for a long time. You came out of stealth, you announced you have a product. Now you’re a public company CEO at a remarkably young age, for a lot of people, that’s the whole startup journey. You did the whole thing. Is that done for you? Do you see another stage of this journey that isn’t... Like most startup founders when they get here, they start to get antsy, but I’m wondering if you’re feeling that or not.

I think it’s a good, and reasonable question, for any founder along the way. And I think there’s some folks that scale better than others at different parts of the journey, you have different types of personalities along the way. A lot of times early on founders, CEOs, will go and take a more technologically or R&D-focused role along the way or not really handle the business side of the equation with this. At least for me personally, the way that I’ve always seen it is that you have to have a system-level technological solution as well as a really strong business case and mentality if you want to be able to see this build and successfully scale.  A lot of the great entrepreneurs of our era, the Zuckerbergs of this world, or even the Bezoses, took their companies up to this point and scaled them incredibly well.

And I think that’s where... I have a similar mentality and drive and desire to be able to build ultimately to that level and scale of this kind of company. And I mean, certainly if there’s any industry that’s ripe for disruption, has the ability to enable this as a trillion-dollar space, and is evolving, it is autonomy, and that’s where absolutely we’re going to be continuing to drive towards. So it’s a good question.

The most important thing is you bring in the key and relevant top-notch, A-plus team members on board to be able to successfully execute. You can have the best vision on the planet, and just the best strategy, the best product, the best commercial landscape, but you need to execute still. We still have a lot of stuff to execute on, and that’s where we make sure to get just rock star team members from across the board to be able to do exactly that. And whereas a lot of entrepreneurs, maybe you see [an] IPO as like the end of the journey, you’re talking about phase one, phase two, all of this. All of this was our phase one. Now we’re onto phase two. So that’s where it really is the beginning of a long journey. Being on the younger side, I’m in a fortunate position a lot of times.

I think when talking about different public companies out there, a lot of times you have CEOs that only plan to have a tenure on the order of like five years or whatever it may be. I think [I’m] in a fortunate enough position — I’m probably the youngest public company CEO — to be able to have a time horizon and a vision more on the order of 50 years than five years. As we see this through, I don’t plan on, frankly, doing anything different. Now the scope of what Luminar will do, we will expand over time, but we have the time. And you just take it step by step and be absolutely committed 100 percent to see it all the way through, until every car has these kinds of autonomous capabilities and is accelerated by way of being powered by Luminar.

I very much look forward to the interview we do when you’re 75 and I’m 90, and I play clips from this interview back at you. That’s going to be a good time. Well, Austin, it’s been terrific to have you on Decoder. I look forward to talking to you much more as time goes on. I personally cannot wait to have a car that will drive into New York City. So hurry up.

Fantastic. Yeah, that’ll be awesome. All right. Well, thanks so much for having me.

<strong>Decoder with Nilay Patel</strong> /

A new podcast from The Verge about big ideas and other problems.

Subscribe now!