Path planning of a mobile robot in a free-space environment using Q-learning

This paper proposes an improved Q-learning algorithm for the path planning of a mobile robot in a free-space environment. Existing Q-learning methods for path planning focus on the mesh routing environment; therefore, new methods must be developed for free-space environments in which robots move continuously. For the free-space environment, we construct fuzzified state variables for dividing the continuous space to avoid the curse of dimensionality. The state variables include the distances to the target point and obstacles and the heading of the robot. Based on the defined state variables, we propose an integrated learning strategy on the basis of the space allocation to accelerate the convergence during the learning process. Simulation experiments show that the path planning of mobile robots can be realized quickly, and the probability of obstacle collisions can be reduced. The results of the experiments also demonstrate the considerable advantages of the proposed learning algorithm compared to two commonly used methods.