Don't Look Now, Toyota's Autonomous Self-Drifting Supras Are Banding Together

It's happened again: that newfangled artificial intelligence technology everyone is raving about has taken yet another fun job from the hands of skilled, erm, workers. This time it's at Toyota, where the company has developed its autonomous drifting Supra sports car to drift in tandem with a second autonomous drifting Supra sports car. Impressive, and it shows how self-driving technology can apply to a complex situation that's not literally just spinning in circles. There's actually a safety upside to all of this tomfoolery, so here's what to expect from future AI-powered smart systems from Toyota.

The Toyota Research Institute (TRI) together with a Stanford Engineering team have spent the past seven years developing new AI-powered technologies with automotive applications. They got one Toyota to drift, and now they've gone further, with two Toyotas able to tandem drift completely autonomously, with no driver inputs inside the car nor delivered remotely. Testing took place at Thunderhill Raceway Park in Willows, California, using two modified Toyota GR Supra sports cars built to Formula Drift spec.

See All 15 Photos15

Toyota's side is said to have worked on developing stable and durable control mechanisms for the lead car in the tandem drift maneuver, while the Stanford team developed the AI computer models for the chase car to keep up without crashing. Both cars have to share a dedicated WiFi network pairing to communicate in real time, including positioning and planned trajectory.

See All 15 Photos15

Computers use sensors to control the vehicle steering, throttle, and braking at a rate of around 50 scans per second. The AI computing uses a technique called Nonlinear Model Predictive Control (NMPC) which outlines each vehicle's objectives represented mathematically into constraints for the driving program; the lead vehicle is instructed to sustain a drift along a desired route while within limits like max steering angle (so, not breaking the car), and the chase vehicle's role is simpler, which is basically do the same thing based on the first vehicle's movements and react in real time.

See All 15 Photos15

So, how does this apply to a safer future for those of us who still drive the car ourselves? “The track conditions can change dramatically over a few minutes when the sun goes down,” said Stanford's Chris Gerdes. “The AI we developed for this project learns from every trip we have taken to the track to handle this variation ... The physics of drifting are actually similar to what a car might experience on snow or ice. What we have learned from this autonomous drifting project has already led to new techniques for controlling automated vehicles safely on ice.”

While Toyota and others have gotten solo cars to autonomously drift in the past, the novel approach here is training the chase car to adapt to the conditions of the lead car, to better simulate real world situations where a vehicle safety system may need to kick in to help a driver to avoid sliding into other motorists, pedestrians, or obstacles in a dynamic manner that goes beyond mimicking a navigational route or rudimentary braking system, but instead actually reacting to the situation at hand with various vehicle inputs. It looks like simple fun, but it's actually another complex step closer to an AI-driven future that actually might have some benefits.