Active Steering Control of Independently Rotating Wheels Based on Multi-agent Deep Reinforcement Learning with Experimental Validation

This research introduces a control algorithm based on multi-agent reinforcement learning (MARL) designed to enhance the running performance of independently rotating wheels (IRW). The algorithm's effectiveness is experimentally validated using a 1:5 scale model. For this purpose, we constructed a dynamic simulation model of the scaled prototype and developed a training loop for a model-free controller. The front and rear wheelsets are considered individual agents, and the multi-agent deep deterministic policy gradient (MADDPG) algorithm is applied to realize distributed steering control. The proposed controller is implemented on an embedded computer within the scaled vehicle on a scaled experimental track, where the lateral displacement and attack angle of the wheelset are calculated through 3D laser profiler sensors. The experimental results show that the steering performance of IRWs on both straight and curved tracks is significantly improved when using the MADDPG controller.