Autonomous Drifting AI Emerges for Safer Winter Driving on Ice

A team led by Elliot Weiss at Stanford University has demonstrated that autonomous vehicles can use controlled drifting to avoid collisions on icy roads—a capability that emerges naturally at higher speeds rather than being explicitly programmed. Their paper, accepted at IEEE ITSC 2026, presents a drift-capable nonlinear model predictive control (MPC) system designed for scenarios grounded in real-world crash fatality data. The controller was tested in high-fidelity simulations across two dangerous situations: hitting a patch of ice on the rear axle during a road departure, and avoiding an oncoming vehicle that has slid into the lane. In both cases, the MPC system autonomously initiated and sustained drifting maneuvers to stay on the road or evade obstacles.

Compared to a benchmark electronic stability control (ESC) system—which prioritizes stability by preventing sideslip—the drift-capable controller deliberately traded stability for controllability, allowing it to precisely maneuver through dangerous winter driving scenarios. A Monte Carlo study over random ice patches showed that the drifting controller achieved lower median lane error than ESC at several speeds, with drifting predominantly emerging at speeds above 50 mph. The results challenge conventional autonomous vehicle safety systems that avoid slipping at all costs, suggesting that controlled loss of traction can be the optimal response in icy conditions. This work brings practical AI-driven drifting closer to real-world deployment, potentially reducing the 400+ annual winter driving fatalities in the U.S. alone.