TANDEM DRIFTING ON AUTOPILOT
Lessons learned from complex maneuvers on the racetrack could lead to improved safety for the average driver.
Written by Mark Wolverton
FOR MOST FOLKS, THE SUBJECT OF AUTOMOTIVE SAFETY is probably pretty dull, evoking memories of driving safety films back in high school and lectures about safety belts. So, what was a group of engineers from Stanford Engineering and Toyota Research Institute (TRI) doing on a racetrack in California, pulling spectacular driving stunts straight out of Vin Diesel movies?
“Yeah, that’s a good question,” laughed Chris Gerdes, professor of mechanical engineering and co-director of the Center for Automotive Research at Stanford University. The answer, it turns out, is that there’s a lot to learn from the most extreme and daring drivers and their cars that can make driving a lot safer for the average person just heading down to the supermarket—especially from something known as “tandem drifting.”
In this drifting maneuver, the driver deliberately spins the rear wheels of the car to make them break traction with the road, allowing the car to slip or “drift” as desired around a turn. First devised by European race drivers in the 1950s, drifting later developed into a competitive motorsport in Japan and then the United States, including variations such as tandem drifting, in which a lead car’s moves are closely mirrored by a chase car. It’s a hair-raising maneuver in which the cars are moving while mere feet or even inches apart at high speed, requiring the utmost skill and daring to perform moves that conventional vehicles never do—at least, not intentionally.
On a snowy or icy road, even the most careful driver might find their car sliding or spinning out of control. Can safety systems be built to reduce the danger for conventional or even fully automated cars? That’s the question Gerdes has been exploring.
“We’ve been very interested in racing and what we can learn from race car drivers,” he said. “We’ve been trying to learn how drivers control cars because the best human drivers are phenomenal.”
Above: Stanford Engineering researchers, in collaboration with the Toyota Research Institute, have achieved the world’s first fully autonomous tandem drift, with the goal of advancing the potential of AI to improve safety. Photo: Toyota Research Institute and Stanford Engineering
“As we’ve explored racing and drifting and looking for things that might be a parallel for safety out on the roadway, we began to be really fascinated by formula drift and what those drivers can accomplish.”
Chris Gerdes, co-director of the Center for Automotive Research, Stanford University
A precise pas de deux
Since around 2009, Gerdes and his team have been working with automated race cars.
“As we’ve explored racing and drifting and looking for things that might be a parallel for safety out on the roadway, we began to be really fascinated by formula drift and what those drivers can accomplish,” he explained. “They’re able to do these things so incredibly smoothly, so we looked at them originally as the benchmark of what automated vehicles should be able to do if we’re going to avoid crashes. We started working with some of the drivers and with Toyota Research Institute to try to figure out what we learn from them and actually get the technology to that level of performance.”
As it happens, the physics governing the movement of a drifting race car and the family SUV sliding on an icy road are quite similar. The Stanford team already demonstrated single-car drifting in an autonomous DeLorean in 2015, and TRI did likewise in 2022. But tandem drifting is the most complex challenge in motorsports.
Working with the engineers at TRI, Gerdes and his crew set out to build an AI system that could fully control two autonomous vehicles in tandem drift. Two GR Supras were specially modified with increased horsepower, computer control of steering, throttle, and brakes, and an array of sensors to keep track of velocity, position, and rotation. Each car was set up according to the official specifications of Formula Drift competition, including its engine, tires, suspension, and transmission.
Although each car carried a driver aboard, each was controlled strictly by AI algorithms, developed by TRI for the lead car and by Stanford for the chase car.
Two vehicles tandem drifting, autonomously. Photo: Toyota Research Institute
Beyond the track
“We’re trying to take advantage of the computational and sensing capabilities of the vehicle, but then from a strategic standpoint, we actually are in some ways emulating what the human drivers do,” Gerdes said.
Most critically, the cars were linked by a dedicated WiFi system, allowing them to communicate with each other and closely coordinate their movements through a technique called Nonlinear Model Predictive Control (NMPC).
With NMPC, each car is given a set of instructions or objectives, represented mathematically: The lead car drifts in a desired path according to Formula Drift rules and the limitations of the vehicle, while the chase vehicle drifts along in tandem. The NMPC algorithms in each car solve optimization problems up to 50 times a second to decide on the proper brake, steering, and throttle commands to apply. About 12 to 15 successful tests were conducted at Thunderhill Raceway Park in Willows, Calif.
Doing autonomous drifting with two souped-up sports cars in a controlled setting is all very cool, but Gerdes and his team are looking to translate this work into improved safety for the average driver. By adapting and translating the lessons learned from exotic stunt driving to stability and safety control systems for production cars, the researchers hope to sharply reduce the more than 40,000 yearly U.S. traffic fatalities.
“The idea is that this is kind of a fundamental learning that would work in future safety systems for consumer vehicles,” Gerdes said.
Both research teams have tested the system in real-world conditions by taking the same software used for drifting and applying it to situations such as an icy road. “[Our system] takes the understanding of the physics of drifting, and it finds some interesting things to do on ice that you can do with production hardware that actually work to increase safety,” he explained. “What we’re doing now is looking through all of this data that we have, which will really point toward the next steps.”
Mark Wolverton is a technology writer in Narbeth, Pa.
© 2025 The American Society of Mechanical Engineers. All rights reserved.