Reinforcement Learning-Based Path Planning for Autonomous Robots ∗

In this paper a path planning method, based on reinforcement learning, is proposed. It was implemented for a small two-wheeled robot. This approach is inspired on potential field functions and reinforcement learning. The basic idea is to model the problem as a state-action problem and find a function that give us the value of each state, which means how good is to stay on it, in the workspace. Starting in a state and following in the direction of the gradient of the function, a robot is going to arrive to the goal state. The solution given by this method is easier to be executed by a robot with non-holonomic constraints than the solutions of the most commons methods, which generate almost always a polygonal line path.