Roborace, Steered by AI, Zooms Towards the Starting Line

by Tony Kontzer

As auto racing gets reinvented, AI is playing a starring role.

At the center of this evolution is Roborace, which is set to launch a driverless Formula 1-style road-racing series. Auto buffs in attendance at the GPU Technology Conference got a glimpse at what’s under the hood of the powerful autonomous cars the company has developed. And they were impressed.

Conceived by renowned car designer Daniel Simon — a former Bugatti designer who’s gone on to create various cars for Hollywood — the so-called “Robocar” features 4 electric motors, 15 sensors, 6 computer vision cameras, and front and rear radars. It’s all controlled by the NVIDIA DRIVE PX 2 AI computer for self-driving cars. The cars — which are 4.8-meters-long — can reach speeds of over 300 kilometers per hour.

Impressive as those specs are, they have little to do with how races will play out. Roborace is designed to highlight the driving capabilities of AI, not mechanical prowess. Driving teams — expected to include equipment OEMs, universities, mobility companies, and even traditional racing teams — won’t be able to soup up their vehicles with additional sensors or more horsepower. Instead, they’ll be judged based on the quality of their algorithms and their AI capabilities.

“What we don’t want this to be is a hardware competition,” said Roborace CTO Bryn Balcombe, in a GTC session.  “The AI drivers that sit inside these cars are the key performance differentiators.”

That means every car will have the same power under the hood. They’ll also have the same assortment of sensors and cameras and radars feeding them the same flow of information.

“The key,” said Balcombe, “is what you do with that information.”

Roborace introduced itself to the world at last year’s GTC, and a year later it has three prototypes of its first vehicle, with a second vehicle design expected later this year. By the end of 2017, Balcombe said the company hopes to have two cars conducting test races on a track.

Much of Roborace’s R&D is conducted in a simulation environment it built. The resulting AI technology is tested and validated in another race car, with a driver on board to ensure safety. Eventually, the racing series will feature 10 teams.

In the meantime, Balcombe said the focus is on refining the vehicles to maximize their performance. That means looking hard at everything, including how AI manages the physical parts, how tactical decisions such as whether to pass a competitor are made, and how the onboard systems process instructions and advice from the teams overseeing the vehicle. Roborace also will factor in accidents, asking the AI drivers to assess the odds that a given maneuver will result in contact.

“We’re pushing the skills of these AI drivers,” says Balcombe. “Anything you can control with software can become a performance parameter.”