A Reinforcement Learning Enhanced Fuzzy Control for Real-Time Off-Road Traction System

In deformable terrain conditions, the tire-surface gripping may be characterized by drastic, rapid, and frequent changes, and, thus, the response time of traction control systems (TCS) of off-road vehicles is crucial for real-time mobility improvements. To advance TCS performance, the time boundaries for the TCS response time is established in the paper based on a physical property of the transient tire traction force, which is the tire relaxation time constant. A Q-learning (QL) algorithm is synthesized to ensure the established time boundaries and to provide real-time TCS response. Then, the reinforcement learning algorithm is proposed to adjust parameters of a fuzzy logic controller (FLC). The proposed control method outperforms the straightforward reinforcement learning in terms of the smoothness of the output signal and control action while also ensuring the established time boundaries. The mathematical modeling and simulation study is applied to an open-link locomotion module described in the paper.