Path Planning for a Mobile Robot in Unknown Dynamic Environments Using Integrated Environment Representation and Reinforcement Learning

This study develops a new path planning method which utilizes integrated environment representation and reinforcement learning to control a mobile robot with non-holonomic constraints in unknown dynamic environments. With the control algorithm presented, no approximating the shapes of the obstacles or even any information about the obstacles' velocities is needed. Our novel approach enables to find the optimal path to the target efficiently and avoid collisions in a cluttered environment with steady and moving obstacles. We carry out extensive computer simulations to show the outstanding performance of our approach.