- Ben Foxall is interested in how humans and tech work together.
- He began working for Wayve, a startup that's using AI to teach cars to drive.
This as-told-to essay is based on a conversation with Ben Foxall, a data engineer in London. It has been edited for length and clarity.
I'm a data engineer at Wayve, a self-driving-car AI company, where I run the robot web team. My team oversees the whole process of how self-driving cars work. We create interfaces — platforms where we can see the AI software and the car interacting — to see what's happening in Wayve's robot cars.
I sometimes describe our user interface as a "window into the robot." It allows us to see what's going on with the model operating the car. It lets us see what's happening in the vehicle's cameras and sensors, as well as its speed, steering, indicating, and braking functions.
We also have an outward-facing user interface, which is essentially the tablet that passengers can see on the dashboard when they get into a Wayve vehicle. This shows the basic details a passenger would need to know, such as where the car is at any given time.
Models created by our research scientists control our cars
Our models use a specific approach to self-driving technology called AV2.0 that can drive using end-to-end deep learning.
Our technology uses a lot of reinforcement learning, taking examples from expert safety operators, virtual data generated in our in-house simulator, and real-world testing experience.
With reinforcement learning, you experience something once and learn what to do. We're not physically coding preprogrammed rules into the car that specifically tell it what to do.
It's similar to how you or I would learn how to drive from experience and the "data" we take in while we're driving. If you've experienced what to do at a traffic light, for example, you know what to do at that type of traffic-light scenario in another location.
We build the model and train it based on the data so it develops knowledge of, for example, how to stop at a stoplight.
The models that our applied scientists are generating act as the "driving intelligence" of Wayve's vehicles. When programmed into our vehicles, which is what my team does, it enables them to drive autonomously.
An incident that happened when I was younger drew me to this industry
I had a friend at school who died in a car accident. She was doing a classic maneuver and didn't check her blind spots.
It's tragic that people can die in very preventable situations. But I think there's a significant potential for tech to be safer than humans, because tech can perceive a larger amount of data and doesn't get tired.
Autonomous-vehicle safety operators have to be in our cars when we're testing a model out on the road. They train for three months and are prepared to intervene if needed. Because the model is learning from behavior, if a safety operator has to intervene, it will know not to do that again.
I feel like people are getting used to the idea of autonomous cars
When I first started working in self-driving technology, people would say to me: "Is it safe? Aren't you scared of it?" But over the last few years it's changed to "When is it going to happen?"
Wayve's headquarters is in King's Cross, one of the busiest areas of central London, so it's a really challenging environment to drive in. The first time I rode in one of Wayve's autonomous cars driving around the area, I was incredibly impressed to see how our models handled it. Honestly, there are times you forget you're in a vehicle powered by AI.
Right now we're building commercial fleets of self-driving cars for urban areas to be able to take over the last-mile delivery of groceries. We've already started trials with our partners at Asda and Ocado Group, two major UK grocery chains.
Humanity's relationship with technology is constantly changing. My favorite example is light switches. When electricity first came around, they were this kind of mystical, magical thing. But over the years they've become a thing we trust in, even though how they work is very complex.
Self-driving cars are mind-bogglingly complex and incredible. But being a passenger in an autonomous car will be like flipping a light switch in the future.