Powerslide Control with Deep Reinforcement Learning

Abstract

Controlling a vehicle’s powerslide motion in the presence of a human driver is a challenging control task, but one that may have a significant impact on vehicle safety, for example, during rapid evasive manoeuvres. Reinforcement Learning, a data-driven optimal control strategy, has gained increasing attention in recent years, demonstrating its effectiveness in successfully controlling various nonlinear systems. In this work, a novel powerslide controller is designed for an all-wheel drive battery electric vehicle with individually driven front and rear axles and a human driver in closed-loop using Reinforcement Learning. The performance of the proposed controller is analysed, and its robustness to steering disturbances and changes in road friction is demonstrated.