Losing traction while driving at high speed is usually very bad news. Scientists from the Toyota Research Institute and Stanford University have developed a pair of self-driving cars that operate artificial intelligence to do this in a controlled manner — a trick better known as “drifting” — to push the boundaries of autonomous driving.
Two autonomous vehicles performed a daring feat in May, drifting in tandem around Thunderhill Raceway Park in Willows, California. promotion movietwo cars race around the track a few feet apart after their drivers hand over control to them.
Chris GerdesStanford University professor who led the effort tells WIRED that the techniques developed for the feat could ultimately lend a hand future driver-assistance systems. “One of the things we’re wondering is whether we can do the same thing that the best drivers do,” Gerdes says.
Future driver-assistance systems could operate algorithms tested on the California track to intervene when a driver loses control, steering the vehicle out of trouble, much like a stunt driver would. “What we’ve done here could be scaled up to larger problems, like automated driving in urban scenarios,” Gerdes says.
The project is a neat demonstration of rapid autonomy, although autonomous vehicles are still far from perfect. After a decade of promises and hype, taxis now operate without drivers in some constrained situations. But the vehicles are still prone to stalling and may require remote assistance.
Researchers from Toyota and Stanford University 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 properties. They have also developed algorithms that combine advanced mathematical models of tire and track properties with machine learning that helps the cars learn how to master the art of drifting.
Ming Linprofessor at the University of Maryland who studies autonomous driving, says the work is an invigorating advance in helping autonomous cars operate in extreme conditions. “One of the biggest challenges for autonomous vehicles is driving safely on rainy, snowy, or foggy days, or in low-light conditions at night,” she says.
Lin adds that the Toyota–Stanford project shows how essential it is to combine machine learning with physical models in the world. “Although it’s just an early demonstration, it’s clearly heading in the right direction,” he says.
Toyota and Stanford first demonstrated the algorithms that would allow autonomous cars to drift in 2022. Performing the trick with two vehicles at once requires even more control and involves the vehicles communicating with each other. The cars were fed data from laps driven by professional drivers, whose computers calculated an optimization problem up to 50 times a second to decide how to balance the steering, throttle and brakes.
“We really care about how to control the car in extreme conditions when the tires are slipping, which is the kind of condition where… [encounter] when you’re driving on snow or ice,” says Avinash Balachandran, vice president of Human Interactive Driving at TRI. “When it comes to safety, being an average driver just isn’t good enough, so we really want to learn from the best experts.”
The world has seen incredible advances in AI recently, thanks to the immense language models that power programs like ChatGPT. But as the dual drifting demo highlights, mastering the messy, unpredictable physical world remains a very different proposition.
“In LLM, a hallucination might not be the end of the world,” Balachandran says, referring to the way immense language models will misinterpret facts. “Of course, it could be very different in the case of a car.”