A Self-Adaptive Double Q-Backstepping Trajectory Tracking Control Approach Based on Reinforcement Learning for Mobile Robots

When a mobile robot inspects tasks with complex requirements indoors, the traditional backstepping method cannot guarantee accuracy of trajectory, leading to problems such as instrument not being inside image and focus failure when grabs high zoom. In order solve this problem, paper proposes an adaptive based on double Q-learning for tracking controlling trajectory robots. We design incremental model-free algorithm Double-Q learning, which can quickly learn rectify controller gain online. For rectification problem in non-uniform state space exploration, we propose active learning exploration that incorporates memory playback well experience mechanisms achieve online fast agents. To verify feasibility algorithm, perform verification different types trajectories Gazebo physical platforms. The results show control be used controller’s gain. Compared Backstepping-Fractional-Older PID Fuzzy-Backstepping controller, Double Q-backstepping has better robustness, generalization, real-time, stronger anti-disturbance capability.