Centimeters apart: Autonomous vehicles master extreme sideways driving skills

Toyota and Stanford University have been pushing autonomous driving technology beyond the limits of traction for several years now, and now they've released a video of two autonomous vehicles drifting together, going sideways with extreme precision.

Drifting as a form of motorsport has been around for a while. Most, if not all, drift cars today are rear-wheel drive cars with lots of horsepower and stiff suspension. Often, the steering knuckles are modified to turn much tighter than a stock car, so that they can go more sideways without skidding in reverse lock.

The two-wheel drifting style we now know as D1 competition was popularized in the 1970s when “Drift King” Keiichi Tsuchiya practiced his sideways technique on the mountain passes of Japan. It began to become more common in the late 1980s. One of the first international drifting events outside of Japan was held in Willow Springs, Southern California in 1996.

The sport involves deliberately throwing a car sideways (oversteering) by using the handbrake or clutch modulation on the rear brake calipers – all while spinning the rear wheels with controlled throttle inputs. In competition, drivers are judged on how fast they can go, how far they can go, how much tyre smoke they can produce… and basically how cool they look when going through a series of corners.

Tandem drifting two cars is the same – with the added danger that the following car will try to maintain the same drift angle as the car in front, without crashing into it. Sometimes they are mere millimetres away from touching each other, and often paint is swapped in the heat of the moment.

This week, Toyota and Stanford University announced their latest achievement: the world’s first tandem autonomous drift cars. This incredible feat comes after nearly seven years of collaborative research and development between the two organizations.

The significance of this breakthrough goes far beyond the undeniable wow factor. Drifting closely mimics the challenges of driving on slippery surfaces like snow or ice. Toyota and Stanford have created a testbed for dynamic scenarios by introducing a second vehicle that must precisely interpret and react to the movements of the vehicle in front.

This setup simulates real-world conditions where autonomous vehicles must respond to unpredictable elements like pedestrians, other vehicles, and sudden obstacles, and has the potential to revolutionize the safety and adaptability of autonomous driving technology. The idea is to make autonomous cars as comfortable as skilled human drift racers beyond the limits of traction, so they have elite-level skills at skidding, correcting, and controlling themselves in dangerous and slippery conditions.

Thunderhill Racecourse Park in Northern California provided an excellent testing and proving ground for such a project.

“The physics of drifting are actually similar to what a car might experience on snow or ice. What we learned from this autonomous drifting project has led to new techniques for safely controlling autonomous vehicles on ice,” says Chris Gerdes, professor of mechanical engineering and co-director of the Center for Automotive Research at Stanford.

This isn’t the first time Stanford has developed an autonomous drift car. Back in 2015, it had a fully functional autonomous drift car made from a DeLorean and converted to an electric drivetrain, sadly lacking a flux capacitor. I’ve seen this machine on the track maybe a dozen times, from its first day on the drift platform, where it sat motionless for most of the day with coders gathered around it, to where it could confidently swing autonomous donuts like a Mustang owner by week two.

About two and a half years ago, Toyota got the Supra to slingshot around the corners at Thunderhill Raceway West. Joining forces with Stanford and the Toyota Research Institute (TRI) took things to a whole new level of technical sophistication.

“Our researchers came together with one goal in mind: How can we make driving safer,” said Avinash Balachandran, vice president of TRI’s Human Interactive Driving division. “When your car starts to skid or slide, you rely solely on your driving skills to avoid colliding with another vehicle, tree or obstacle. The average driver struggles to cope with these extreme conditions, and a fraction of a second can mean the difference between life and death. This new technology can intervene just in time to protect a driver and manage the loss of control, just as an expert drifter would.”

It's worth noting that both vehicles are similarly modified Toyota GR Supras, but the lead Supra has been programmed by TRI developers to be stable and repeatable in order to make for safe driving.

The Chase Supra’s AI neural network was developed by Stanford Engineering with a focus on dynamically adapting to the car in front of it so it can drift alongside it without crashing into it. It can learn and improve with each track outing.

Both cars are built to the same specifications required in Formula Drift, but with the addition of additional computers and sensors to collect and share data – such as relative position, speed, turn rate and planned trajectories across a private WiFi network. The computers allow the cars to continuously read and make brake, throttle and steering adjustments up to 50 times per second while drifting, using a technique called Nonlinear Model Predictive Control (NMPC).

For many of us who spend time driving in snowy, icy, or slippery conditions, maybe a little AI guidance could help steer us in the right direction rather than skidding off the road… In your car, of course. Not in mine, thanks.

You can watch Toyota's video here:

Source: Toyota Research Institute