Path tracking control of non-holonomic car-like robot with reinforcement learning

Abstract

This paper investigates the use of reinforcement learning in solving the path-tracking problem for car-like robots. The reinforcement learner uses a case-based function approximator, to extend the standard reinforcement learning paradigm to handle continuous states. The learned controller performs comparable to the best traditional control functions in both simulation and also in practical driving.