Losing traction while driving at high speed is generally very bad news. Scientists from the Toyota Research Institute and Stanford University have developed a pair of self-driving cars that use artificial intelligence to do this in a controlled manner — a trick known as “drifting” — in a bid to push the boundaries of autonomous driving.
The two autonomous vehicles performed the daredevil stunt of tandem drifting around Thunderhill Raceway Park in Willows, California, in May. In a promotional video, the two cars race around the track just feet apart after human drivers give up control.
Chris Gerdes, a Stanford University professor who led the project's involvement, tells WIRED that the techniques developed for the feat could ultimately help future driver-assist systems. “One of the things we're looking at is can we do as well as the very best human drivers,” Gerdes says.
Future driver-assistance systems could use the algorithms tested on the California track to intervene when a driver loses control, steering a vehicle out of trouble like a stunt driver would. “What we’ve done here can be scaled up to address larger problems, like automated driving in urban scenarios,” Gerdes says.
The project is a great demonstration of high-speed autonomy, although self-driving vehicles are far from perfect. After a decade of promise and hype, taxis are now operating driverless in some limited situations. However, the vehicles still get stuck easily and may require remote assistance.
The Toyota and Stanford University researchers have modified two GR Supra sports cars with computers and sensors that track the road and other vehicles, as well as the cars' suspension and other characteristics. They have also developed algorithms that combine advanced mathematical models of tire and track characteristics with machine learning to teach the cars how to master the art of drifting.
Ming Lin, a professor at the University of Maryland who studies autonomous driving, says the work is an exciting advance in helping self-driving cars operate at the extremes. “One of the biggest challenges for autonomous vehicles is operating safely on rainy, snowy or foggy days, or in poor lighting at night,” she says.
Lin adds that the Toyota-Stanford project demonstrates the importance of combining machine learning with physical models in the world. “Although it's still an early demonstration, it's clearly moving in the right direction,” she says.
Toyota and Stanford first demonstrated algorithms that could allow autonomous cars to drift in 2022. Getting two vehicles to perform the trick simultaneously requires even greater control and communication between the vehicles. The cars were fed data from laps driven by professional racers, and their respective computers calculated an optimization problem up to 50 times per second to determine how to balance the steering, throttle and brakes.
“What we're really looking at here is how you can control the car at extreme performance, when the tyres are slipping, the kind of conditions you would be driving in. [encounter] “When you’re driving on snow or ice,” said Avinash Balachandran, vice president of TRI’s Human Interactive Driving division. “When it comes to safety, an average driver just isn’t good enough, and that’s why we really want to learn from the best experts.”
The world has seen remarkable advances in AI of late thanks to the large language models that power programs like ChatGPT. However, as the dual drifting demo highlights, mastering the messy, unpredictable physical world remains a different story.
“In an LLM, a hallucination might not be the end of the world,” Balachandran says of the way large language models misrepresent facts. “That might be very different in a car, of course.”
