A Global Path Planning Algorithm for Robots Using Reinforcement Learning

Path planning is the key technology for autonomous mobile robots. In view of the shortage of paths found by traditional best first search (BFS) and rapidly-exploring random trees (RRT) algorithm which are not short and smooth enough for robot navigation, a new global planning algorithm combined with reinforcement learning is presented for robots. In our algorithm, a path graph is established firstly, in which the paths collided with the obstacles are removed directly. Then a collision-free path will be found by Q-Learning from starting point to the goal. The experiment results illustrate that it can generate shorter and smoother paths, compared with the BFS and RRT algorithm.