Kernel-Based Representation Policy Iteration with Applications to Optimal Path Tracking of Wheeled Mobile Robots

Abstract

AbstractHow to improve the generalization and approximation ability in reinforcement learning (RL) is still an open issue in recent years. Aiming at this problem, this paper presents a novel kernel-based representation policy iteration (KRPI) method for reinforcement learning in optimal path tracking of mobile robots. In the proposed method, the kernel trick is employed to map the original state space into a high-dimensional feature space and the Laplacian operator in the feature space is obtained by minimizing an objective function of optimal embedding. In the experiments, the KRPI-based PD controller was applied to the optimal path tracking problem of a wheeled mobile robot. It is demonstrated that the proposed method can obtain better near-optimal control policies than previous approaches.KeywordsReinforcement learningpath trackingkernel methodswheeled mobile robotsapproximation policy iteration