Earlier this year, Roborace challenged two universities -- the Technical University of Munich and the University of Pisa -- to develop a faster car in Berlin. Both teams drove the Devbot, a test car with a cockpit, six times around a 2-kilometer track. The first three were with a human behind the wheel, and the final three were with software alone. The team with the fastest average lap was crowned the "Human and Machine Challenge" winner. "It was very interesting," Balcombe said, "because it was the first time we didn't use our software."
Robocars are expensive, so the company has to figure out how, if at all, it should be restricting new teams. The company would like to avoid engineers, after all, flipping the car on the first corner of every race. In Berlin, the two universities were forced to pass a makeshift driving test before they could race in a competitive environment. In traditional motorsport, drivers start in slower go-karts and earn licenses before competing in faster, more dangerous forms like Formula 1. Roborace is thinking about a similar licensing system for engineers who want their algorithms to compete in its competitions.
"At what level is this AI driver?" Balcombe said. "Is it near that go-karting starting point, or is it Formula 1 level? That's a big spread in capability."
The company is talking to cities, too, that might want to host one of its races. Roborace is pitching itself as the pinnacle of autonomous technology, which could appeal to mayors who want to show off their town's smart and connected credentials. Hosting a Roborace could also be an effective way to attract teams and, by extension, nurture local experts in the field. "There's a global competition at the moment for talent," Balcombe said, "and if you can showcase that your city is the best place to develop this technology then you kind of want an event to showcase that."
At the same time, Roborace needs to show up at public events and dazzle visitors to maintain interest.

Bryn Balcombe, Chief Strategy Officer for RoboraceCharging through chicanes is no easy task, however. Roborace spent six months preparing for its Goodwood hill climb. It involved some human reconnaissance and a practice run with the Devbot, which produced an accurate, three-dimensional map for the Robocar to reference.
The trees around the course meant that GPS wasn't an option. Roborace encountered a similar problem during a demonstration in Hong Kong last year. Densely-packed skyscrapers, known as street canyons, regularly block out parts of the sky. To compensate, the team had to develop an entirely new software platform based on LIDAR, the laser-based equivalent of radar (most self-driving cars use LIDAR to navigate and spot potential hazards on the road.)
"When the car left I thought, 'they've got that set too fast.'"
The solution Roborace developed in Hong Kong, though, is unable to differentiate grass and tarmac. To run the car at Goodwood, then, it needed a set of machine-vision cameras that could discern driveable surfaces and pipe this information into the mapping and navigation systems. Finally, the team set up a course at Magny-Cours in rural France that reflected the hill climb at Goodwood. Robocar was driven at speeds higher than 120 KMH -- the vehicle has reached 210 KMH (roughly 130 MPH) before in private testing -- but capped in southern England for safety reasons.
Before the demonstration on June 13th, Roborace conducted a practice run with the Robocar. "It was incredible really," Balcombe said. "I was standing at the start line and when the car left I thought, 'They've got that set too fast, surely they've got that set too fast.' The dust that it was kicking up, I was like 'Oh my God!'" It wasn't too fast, however. The car made it to the top of the course just fine, settling any remaining nerves about the public showcase.