Robust Reinforcement Learning-based Vehicle Control with Object Avoidance

Abstract

Listen

Translate article

This paper presents a control design framework for robust reinforcement learning (RL) based vehicle control enhanced with object avoidance using a model predictive approach. The proposed integration method is applied for motion control of autonomous road vehicles. In the motion control, longitudinal and lateral dynamics are incorporated, and the high-performance motion of the vehicle, through the RL-based control agent is achieved. ’H.co based robust control is used as a baseline parallel to the RL-agent to ensure stable operation. The object avoidance and limited error path following are achieved using a model-based predictive algorithm. Finally, the control signals of the three controllers are combined using a supervisor layer, which solves a constrained optimization task to ensure high performance and safe motion. The effectiveness of the proposed control method using simulation scenarios is illustrated.