View from Initialized

The future of practical robotaxis is here

Autonomous car with no safety driver for $50M, not $10B

You’ve heard of Waymo and Cruise (Initialized was actually an early seed investor in Cruise, too). Each of those teams has spent more than $10 billion to deliver something that can drive normal roads without a safety driver. But there’s another way, and I’m excited to share that with you today. It’s by a scrappy but brilliant startup called Voyage. They’re announcing their third gen car that will operate with no safety driver at all, but having only raised $50M instead. How did this founder and his team do it? Let’s find out.

Garry: Oliver, thank you so much for joining us today. You’ve built world-class autonomous car company and you’ve done it for only $52 million.

Oliver: We recently announced our third generation robotaxi. We call it the G3 and this will be a first vehicle that is operating without the need for a safety driver in the front seat. And for us, this is a very big deal because this is the culmination of four years of work and a collaboration with Fiat Chrysler. So what you see here is one of our robotaxis driving by itself through one of our communities. This is a retirement community in San Jose and here we serve senior citizens. We move them from their doorstep to any other address within this particular community. And here we serve a whole bunch of different applications: We enable seniors to go visit friends, enable them to go to the park, enable them to go to many of the points of interest in this particular community–eight community centers, swimming pools, gyms, stables, all of these very different things. And what you see right here is our technology in action.

There are three components in play here. There is what we call Commander, Shield and Teleassist. Commander really is the brain of our robotaxi. It’s responsible for perceiving the world, predicting the world and ultimately planning a safe path through the world. And then we have Teleassist. And Teleassist is a very unique take on remote assistance. What we very much believe is that robotaxis today . . . it’s an impossible task to have a robotaxi perform to the same level of intelligence that a human has for certain edge cases. Think about wild animals roaming around your vehicle. We’ve had numerous times where wild turkeys have jumped out in front of our vehicle, our vehicle comes to a stop and then they just don’t move. Now a robotaxi has no idea how to deal with that right? It doesn’t have human intelligence. What do we do as humans? Well, we honk the horn, we kind of nudge and drive a little bit towards them, and then they flee right? And what we’ve built with Teleassist is a very safe way for a remote human, one of our what we call Teleassistants to directly and remotely control the vehicle and do so very safely. And that enables us effectively to handle any and all edge cases that the world could throw at us without having to devote years and years to research and development and to experience all these edge cases.

And then we have Shield, which is a reliable backup system. You can think of Shield as an extra set of eyes to hit the brakes if it believes the vehicle needs to come to a stop. Our whole self-driving technology is designed with redundancy in mind. We assume that everything is going to fail and we design software and hardware to mitigate the impacts of those failures so we can handle it very safely and Shield is a very crucial part of that.

Power and Commander are a number of state-of-the-art algorithms in a variety of different areas. State-of-the-art algorithms in areas like 3D object detection, semantic segmentation, stereo vision, stateful decision-making. And all of these algorithms have been finely tuned and hardened so that we have a very reliable vehicle that can take people from point A to point B. And this is the technology we’ve been building now for nearly four years.

Garry: Really awesome stuff. And it’s all actually happening live in vehicles every single day in your first communities.

Oliver: That’s exactly right. This vehicle has been on the road now for four years. And in these places, we’re serving a number of different customer requests, moving people from their doorsteps to wherever they want to go.

Garry: As they say, the future is already here, it’s just not evenly distributed.

Oliver: Exactly. Your grandma is going to ride in a self-driving car before many of us around the world. I think that’s very apt in a startup sort of sense, right? It’s that you typically don’t want to build a product for someone who already has a pretty good solution. And if you look at humans driving cars today, non seniors driving cars today, we’re relatively okay and pretty good at it. When you look at the senior category, again, there’s so many things that just stand out as clear improvements to the experience and to the safety of the current mode of transportation. The fact that a senior is 17x more likely to suffer a fatality is just mind boggling, right? We can make a dent in that number by providing a vehicle that is ultra safe, has superhuman vision, can see in 360 degrees day or night, that’s going to make a big difference in the life of that customer, but perhaps not in the life of a customer today who is able to drive their own car.

Garry: I mean, I love this simply because it fits my fundamental view that startups that become billion dollar startups actually have a very simple premise that ends up being better, faster, cheaper for their customers. And in this case faster by a lot, especially once you get to a point where you don’t need a safety driver, you can have a fleet at the ready with really great support for the demand of a given community. Better in that it’s way safer clearly. And then cheaper because that’s sort of long-term where a lot of this goes. How have you been able to outperform a lot of those companies?

Oliver: There is another way. And the way we’ve always approached this problem is to think like a startup, right? To really embrace constraints and embrace that we cannot build everything. We can build the things that we think we can be really really, really good at. And for us that looks like the core self-driving technology. It’s perception of the world, it’s the prediction of the world, it’s the planning of the world. And we believe we can do a really good job with that resulting in a great product for the customers that we look to serve. But what is absolutely crucial is that from day one you don’t reinvent the wheel, you don’t try and build everything, you don’t build your own sensors, you don’t build the things that the ecosystem can provide you. And because of that approach, we’ve made really fast progress toward delivering a robotaxi for a customer that really needs it.

Garry: One of the things that’s clear from us working with so many startups is that necessity is the mother of invention. And then from certainly lots of other places in the startup ecosystem whether it’s WeWork as a cautionary tale or things like Magic Leap, you know you can get kind of lost with that much capital. What’s happening there?

Oliver: My general theory here is that the autonomous vehicle industry did not begin like most industries. The very first startup to build an autonomous vehicle was Google, right? And Google was not a startup at the time. It was in fact a very big company that had a ton of money that they could throw at what seemed to be a massive problem. And then what eventually occurred is that everyone looked at how Google was doing it and said, well, we want a piece of that pie too. Let’s go take that same approach with a few deviations on the ideas, but let’s take that same approach, the same need for capital, the same talent and let’s just go effectively replicate that. And that approach has led to companies building their own lidars, their own radars, their own cameras, in some cases designing their own chips, their own cars, their own simulation infrastructure, their own mapping infrastructure. The list is quite endless. (We at Voyage embrace cloud compute). There’s been companies I know that have built their own training infrastructure internally to improve their perception models and they have engineers dedicated to those problems. When you try and take on all those very big challenges, suddenly the self-driving software piece is no longer 100% of the problem–it’s maybe 50% of the problem. And all of these additions that have also very complex problems tend to take away from the time of many of your top people.

Garry: I guess none of them has tried to reinvent the GPU yet though. It sounds like maybe that’s a bridge too far that illustrates the conundrum when you’re trying to build something brand new. You might end up spending your own most expensive capital, which is your equity in reinventing things, when frankly, billions of dollars have already gone into other solutions. In the autonomous car ecosystem, I feel like Voyage has been able to benefit greatly from that ecosystem by not succumbing to not-invented-here syndrome, which is quite dangerous for people.

Oliver: The caveat that I should add is that companies like Waymo and Cruise, I think they’re building this like they should, right? They’re building it the way they are because they have some unique strengths. They have lots of capital, and in Cruise’s case, they have a car company standing behind them. Those are very powerful assets. The question then becomes, for a company like Voyage that doesn’t have those unique assets of a Google search engine or a car company, how should you go and do it? Embracing the fact that we cannot and should not solve every problem and that we also can’t solve every technical or customer problem. Robotaxis should be a consumer product and not necessarily a kind of science project. Intentionally from day one, we’ve also tried to limit the set of customer problems we solve. We’re focused today on senior citizens because they have these tremendous transportation pain points. Instead of going very broad and saying, we’re going to replace Uber for everyone, where there’s such a diversity of different challenges. If you raise a bundle of money to go and invest across as many different areas as you can, that’s difficult. It’s quite hard to put money to work in many different ways. If you hire people and if you have an excess of capital, and you go hire more people, they come up with more things to build internally.

Garry: This business is not different than any other business, which is if you can make more money than you spend in a month or quarter or a year, suddenly your default alive is actually profitable. We’re not there yet, obviously, but it’s a pretty wild concept for some of these technology companies. But it’s absolutely something that is a part of your plan.

Oliver: 100%. And we do this a few ways. Although we do partner on the peripheral elements of the industry–again, mapping, simulation, sensors, the vehicle, we partner with Fiat Chrysler for example, on the underlying vehicle platform. We decided from day one to build the self-driving technology and to build three technologies [mentioned above]: Commander, Shield and Teleassist.

You can think of Commander as the brain of our robotaxi. Shield’s a reliable backup system, and Teleassist is our solution for edge cases allowing a human to aid the vehicle in certain contexts. And we built those technologies all in-house because:

  1. We have the talent to go do so. It was in our DNA and that has been very valuable
  2. We want to actually deliver a product that is cost competitive. Ultimately, this is a people moving business, right? And we want to be cost competitive with the primary modes of transportation, which obviously looks like driving a car.

What was really interesting about seniors–and I’d be the first to admit that this was not a part of the master plan–is that not only do they have tremendous pain points when it comes to transportation, but they also pay significantly more to drive and own their own vehicles. A senior pays about 80% more per mile to drive their own vehicle. And that, for us, is pretty magical because instead of competing with the very optimized cost of owning a typical car, a senior pays a lot more. We can undercut that cost far quicker than the others can undercut the cost of normal car ownership. We hope and think that will be this kind of product market fit moment for us. Not only do we solve the driving challenge, but we also solve the cost challenge of owning a car.

Garry: I think that’s one of my favorite things about Voyage . . . You’ve approached this from an extreme engineering standpoint, which is “what are our constraints? How do we optimize against those constraints?” Autonomous cars — there was this idea that eventually there would be level five over any road no matter what, regardless of whether the car has seen it or not. And certainly while I believe that will still eventually happen, you’ve picked a use case that is much more constrained, way more realistic and actually viable right now.

Oliver: The founding DNA is that we didn’t believe we could add much to level five and the kind of science project discussion. We just didn’t feel that we were uniquely suited to add to that discussion. But we were uniquely suited to deliver a product in the short-term that could be very, very valuable to a certain customer and to drive revenue from that product. And then take that revenue and all the momentum we built and expand, expand, expand until one day our company doesn’t look much different than a Waymo or a Cruise. But our beginning just has to be dramatically different than those folks because of all the dynamics we’ve talked about.

Garry: And I think that’s how you ended up making one of the defining autonomous car companies a little bit more like Ford and a little bit less like NASA.

That is a great analogy. Ford started very humbly in a literal garage building, what today is one of the most capital intensive businesses and there wasn’t billions of dollars invested in Henry Ford from day one, right? It was very humble. Again, the NASA approach, if you’ve got the capital then yeah, it’s a worthy attempt right? It’s something that is where certain companies are uniquely suited to doing so. But for us, we’re far closer to Ford than we are to the Apollo missions.

Garry: Oliver this is fantastic. You’ve started multiple companies, taught hundreds of thousands of people how to start their own autonomous driving sort of programs [where you were before]. You’ve become a stalwart in tech today. What are some lessons you wish you knew when you started that you could you impart to the next generation?

Oliver: You’re too kind, but lots of lessons learned, very hard lessons. I remember my very first pitch. In fact, I saw this on one of your videos, Garry, about the pitching Benchmark. My very first pitch was with Roelof Botha from Sequoia back in 2011 and legendary. I think Alfred Lin was in the room as well. I didn’t have a deck, I didn’t know what to do. And it was just a total bomb of a pitch. It was embarrassing, frankly.

Ever since, I’ve been very lucky to learn from a bunch of great people in Silicon Valley about how this should be done. My favorite quote is from one of our board members at Voyage, a gentleman called Sven Strohband. He’s a general partner at Khosla Ventures. Sven basically said, “sometimes you need to survive long enough to be successful because I think all too often everyone’s looking to be instantly successful as if that just magically happens.” There’s a lot of talk in Silicon Valley about overnight successes that in fact take 10 years, right? And it’s absolutely true. And sometimes it’s just about perseverance.

The best companies go through points where I’m sure their founders thought they would die the next day. You just got to keep going and chugging along and surviving. Revenue helps with that, profit helps with that. And the mentality helps with that as well. The best tactical advice I ever got was that you need to answer the question succinctly. You know all too often in fundraising there’s this feeling that you have to prove yourself and you’ve got this chip on your shoulder. And then someone asks you a question and you want to look like the smartest person in the room, right? Like, you know something they don’t. When in reality, put yourself in a VC shoes . . . you’re going through meeting after meeting after meeting and you’re just legitimately asking a question, right? You just want to know the answer. Much of the time in my early time fundraising, just after time in Y Combinator back in 2011, I wasn’t answering the question. Or if I was, it was at the very end of the pitch or the end ramble. You just have to be very to the point on the answer. And if people want more detail, they’ll ask for it.

Garry: Yeah, everything I ever needed to know in tech, I learned in high school newspaper. The inverted pyramid answer is actually extremely useful.

Well, thank you for walking us through this world-changing tech that you’ve built over the past four years. It’s a testament to the team. What you’ve been able to build is something that many other people spent [millions more to do]. I love that because it shows to me that necessity is the mother of invention. Constraints are really one of the big gifts when it comes to creating anything. You’ve done so much with what you have and that bodes very well for the future of Voyage.

Oliver: Fingers crossed, I think it will. And it’s down to the tremendous team we have of nearly 70 folks in Palo Alto, California–coming in every single day and working for our customers and working to build a better way for those folks to move.

Garry: If people want to find out more about Voyage, where can they go? And you are also hiring.

Oliver: The best way to learn more about Voyage is to head to You’ll learn all about what we’re building, our customers, our technology, and of course, yes, we’re hiring. You can see all those open roles and we’d love to talk to you. We don’t put emphasis on degrees and having established backgrounds. We’re very open to folks who are trying to get into this industry in a different way, so please apply. Lastly, I am a relatively prolific tweeter. It’s a little boring, but come check it out, @OliverCameron, and we’ll see you there.

Garry: It’s one of the best resources for anyone who’s interested in computer vision autonomous vehicles. So thank you for tweeting and thank you for appearing here.

Oliver: Thank you for having me, it’s been fun.

To watch this interview, head here.