Application of Deep Q-Learning for Wheel Mobile Robot Navigation

Autonomous mobile robot has tremendous application in various environment due to the fact that they work without human intervention. Path planning and obstacle avoidance are challenging problem for autonomous mobile robot. This paper explores - the obstacle avoidance technique for wheeled mobile robot based on Deep-Q-Learning. In this paper, we introduce a log-based reward value field function which is the reward receives by agent based on relative positions of agent, obstacles and goal. We perform the experiment in simulated environment and physical environment. Finally, we measure the accuracy of the performance of the obstacle avoidance ability of the robot based of hit rate metrices. Our presented method achieves high success rate to avoid collisions.