AI in control of the lead and the chase.
Drifting is hard. Like, really hard. It often takes a ton of repetition, trial-and-error, and refinement to get it right. Consistency is key (just ask Carsen Ruger), especially if you plan on venturing into the high-stakes world of tandem drifting. Incredibly, a group of engineers has figured out how to program two AI systems to tandem drift, throwing the electronic brains into a pair of Toyota Supra race cars and letting them have at it.
The results were captured in a three-and-a-half minute video that shows the autonomous Toyotas doing their thing at Thunderhill Raceway in California. The video includes on-track action, as well as a few insights from the teams that made the world's first autonomous tandem a reality, such as individuals from the Toyota Research Institute (TRI).
"[Tandem drifting] is the most complex maneuver in motorsports, and reaching this milestone with autonomy means we can control cars dynamically at the extremes," said the vice president of TRI's Human Interactive Driving division, Avinash Balachandran. "This has far-reaching implications for building advanced safety systems into future automobiles."
TRI collaborated with Stanford Engineering on the project, with TRI taking responsibility for programming the lead car, focusing on stable, repeatable lead runs, and Stanford Engineering handling the chase car, tailoring the algorithms to adapt to the dynamic motion of the lead vehicle.
Looking at the video, it sure looks like the teams pulled it off. The Toyota Supra race cars were upgraded by GReddy and Toyota Racing Development (TRD) for greater performance, with specs for competition in Formula Drift. Meanwhile, the onboard algorithms constantly optimized while on track, refining the steering, throttle, and brake inputs up to 50 times per second and training the neural network to improve with each lap.
"The track conditions can change dramatically over a few minutes when the sun goes down," said professor of mechanical engineering and co-director of the Center for Automotive Research at Stanford (CARS), Chris Gerdes. "The AI we developed for this project learns from every trip we have taken to the track to handle this variation."
Of course, we prefer to take the wheel ourselves, rather than handing control over to a computer. Still, the fact that the teams managed to pull this off at all is pretty impressive. Up next - AI versus James "The Machine" Deane?
Comments