Reinforcement Learning Based Approach For Mobile Robot Navigation

Abstract

Autonomous navigation of mobile robots has been a subject of research for many decades. Many algorithms, both classical and heuristic have been developed for the navigation of the wheeled mobile robots. Classical approaches can get tedious and can get stuck at local optima as the environment gets more complex. Heuristic approaches are gaining prominence nowadays because of its closeness to human way of behavioral learning. Reinforcement learning (RL) is the idea of having a robot learn how to accomplish a task. It is one of the heuristic approaches which gains knowledge by trial and error, so there is no need for any expert knowledge. In the current work, an attempt has been made to implement reinforcement learning for autonomous navigation of wheeled mobile robots. A RL agent is created for controlling the obstacle avoidance and goal-seeking behavior of a single robot. Then, a kinematic controller is developed to help the robot to move through the waypoints generated by the RL algorithm. Subsequently, the performance analysis of the controller is carried out and the benefits and limitations of the RL algorithm is studied.