Oliver Cameron co-founded the self-driving car startup Voyage in 2017 on the theory that retirement communities are the perfect place to launch a robo-taxi service.
Their low-speed, low-traffic roads make the engineering challenge easier. And retirees, many of whom are unwilling or unable to drive themselves, are an obvious clientele. “We’re focused on serving senior citizens, who we believe are the best first customer for driverless technology,” says Cameron.
Before the pandemic forced it to pause passenger trips, Voyage provided rides in a fleet of ten minivans, with safety drivers, to roughly 4,000 residents of The Villages retirement community in San Jose, California. It also has been testing a small fleet in a community in Florida, coincidentally called The Villages, that is home to 125,000 people. Voyage has multiyear agreements with both locations to be an exclusive robo-taxi provider.
Cameron, a 32-year-old native of Halifax, England, came to the US in 2010 to join the startup accelerator Y Combinator, where he planned to develop smartphone apps that used machine learning. From there he landed at Udacity, the online university founded by Sebastian Thrun, the German computer scientist who also spearheaded Google’s self-driving car project.
At Udacity, Thrun put Cameron in charge of developing a self-driving car curriculum, a project that became the seed of Voyage.
Retirement communities, says Cameron, are only the beginning for Voyage. Eventually the company aims to serve a variety of planned developments as well as towns and small cities, a market that Cameron puts at US$780bil (RM3.23 trillion) annually.
In August, Voyage introduced the third generation of its robo-taxi. Cameron, who has been CEO since the company’s founding, recently spoke with Hyperdrive about the upcoming "G-3” minivan, "tourist” investors, and the path to pulling the safety drivers. The conversation has been edited for clarity and length.
You recently announced the third version of your robo-taxi. What’s notable about it?
The G-3 is our first vehicle that’s ready for driverless operation. It’s built atop a special edition vehicle called the AV-ready Chrysler Pacifica, which has the hardware and firmware modifications necessary for driverless. We integrate our technologies into this. We call them Commander, Shield and Tele-assist. Commander is the brain of our robo-taxi, responsible for making safe, intelligent driving decisions. Shield is our reliable backup system that brings the vehicle to a safe stop if necessary. And tele-assist is our unique take on remote assistance, responsible for getting a customer to that destination even if Commander gets confused.
That means full remote operation?
Yes. What we saw is that there was a long tail of edge cases that you have to experience and then develop the algorithms to handle autonomously. And those things occur relatively infrequently, but they add a ton of time to your development. We're going to solve all those things, but we didn’t want to wait to solve them before delivering our first product. So we decided to build what we call the tele-assist pod, which is a proprietary, custom-built, automotive-grade workstation.
This enables a human to remotely drive the car, at low speeds, around crazy stuff. We’ve seen our vehicle be boxed in by wild turkeys – no joke, about eight turkeys, literally circling the vehicle. A robo-taxi has no idea how to handle that. It doesn't have general intelligence. A human, though, you just honk the horn and maybe start driving a little bit and it will scare them off.
You are using an ultraviolet-C system to disinfect the vehicle between trips. How does this work?
At the end of every trip, the car waits for the passenger to get out and then on its way to pick up the next passenger, for about three to four minutes, that UV-C light is active. Not only does it kill coronavirus on surfaces across all three rows of our vehicle, it also kills other infectious viruses.
Fiat Chrysler is building this AV-ready Pacifica. Nvidia is building the processor for your software. Other partners are making sensors. So you’re not trying to do everything yourselves – you’re stitching the hardware together and creating the mastermind and the business model. Is that a fair way to characterise it?
I think so. We build what we think we can add value to. So we believe we can build a world class self-driving technology, but we don't believe we're going to be the best in the world at building a LIDAR or Silicon chips or vehicles, for sure. Even some of the more nitty-gritty areas like mapping or a simulation engine, we don't see ourselves adding too much value to that conversation, but when it comes to perception, prediction, decision-making, controls, and planning, we think we can add to that equation.
How far are we from widespread driverless vehicles, where humans are not asked to do anything?
The ugly truth is that self-driving technology remains years away from being safe and scalable in dense cities. The state-of-the-art in machine learning and robotics is just not yet up to the challenge. And when it comes to highways and high-speed autonomous trucking, one of the challenges that remains there is fail-operational research and development.
With a truck that is going 65 miles an hour, you’ve got to be damn sure that your system works, that you are not seeing false positives that cause the vehicle to stop when it shouldn't be stopping and cause a pile-up behind it. And what happens when your primary and secondary system, for whatever reason, don’t work? What if your sensors have fallen off?
In a retirement community, where you’re driving 25 miles an hour, you just bring the vehicle to a stop, in lane. It may not be perfect, but it’s perfectly safe. But on a highway you’ve got to keep autonomously driving multiple minutes to get to the side of the road. You've got to cross maybe two or three lanes of traffic.
Has the engineering challenge proved harder than you expected, even within the limitations that you’ve set?
The succinct answer is yes. Even when you try to tailor this technology for a calmer environment, it's still incredibly challenging. That's something that we don't shy away from. The slightly longer answer is that it really depends on the where. I feel very confident that in the next 12-to-24 months, you're going to see self-driving vehicles, with no person in the vehicle, moving people on a daily basis. I would feel very confident in placing a bet there. Now, will that happen in San Francisco? Will that happen in New York City? Probably not. But it will be happening in regions like the regions we serve within 12-to-24 months.
Will Voyage be among those doing it in 12-to-24 months?
I’ve learned from Elon Musk that giving dates specific to our company doesn’t really help anyone. It doesn’t help the company. It doesn’t help external perceptions. My feeling is that the reason the industry feels depressed today is of its own making. It’s been a mismatch of expectations that the industry has brought upon itself. So no dates from us. But we feel confident in our ability to do this. Our challenge is that we have to take the state-of-the-art in self-driving technology and commercialise it. We have to harden it. We have to make it ready for the consumer to be in the car. We don’t need to advance this technology beyond what it's already capable of. – Bloomberg
Did you find this article insightful?