Self-Driving Cars That Can Drift Like Racers Are Coming
What if your future self-driving car could drift around corners like a race pro... safely?
A new study shows drift can emerge naturally from an automated vehicle controller that is simply trying to minimize lap time—no drift reference is ever programmed. The controller, called boundary exploration learning model predictive control, builds a safe set from completed laps and gradually expands the vehicle’s sideslip and yaw rate envelope while staying recoverable. As performance improves, sustained sideslip appears as the rear axle approaches its limit. At a tire-road friction of 0.6, lap time fell from 49.95 seconds on lap 3 to 25.50 seconds on lap 12, with drift first emerging on lap 11 and reaching 16.5 degrees of sideslip. In contrast, no drift appeared at higher friction levels. The authors characterize drift as a conditional continuation of limit handling that emerges when increased performance demand approaches available tire capacity—not as a separate, prescribed motion mode.
- A self-driving car AI can now learn to drift like a pro racer to handle slippery roads better
- In tests, the car cut lap times nearly in half while drifting safely on low-friction surfaces
- This could lead to safer self-driving cars that handle rain, ice, or sharp turns without skidding
Why It Matters
Self-driving cars that master drifting could handle slippery roads safer and faster, saving lives in bad weather.