Explainable DDPG-based Tracking Algorithm for a Steering-capable Robot

A path-following control system for a tracking robot is developed by employing the deep deterministic policy gradient (DDPG) within a reinforcement learning framework. The proposed algorithm is primarily formulated to enable the robot to adapt and learn in complex environments while maintaining a consistent distance from a lead robot. The use of explainable artificial intelligence (AI) techniques for analyzing and visualizing the key parameters of the reward function is emphasized, thereby enhancing trust in AI decisions in DDPG. The effectiveness and reliability of the system and AI judgments are verified through simulations. Results show that the visualization of the decision-making process of the DDPG agent enhanced the comprehension of its actions. Thus, the application of explainable algorithms led to a substantial decrease in the lateral error of the tracking robot.