Path Planning for Unmanned Vehicles Based on Value Function Approximation Algorithm

This paper deals with the path planning problem for unmanned vehicles based on reinforcement learning. Considering the unmanned vehicles' dynamic model, the neural network is used to approximate the value function. Besides, in order to make it more suitable for practical applications and speed up the learning process, the recursive least squares algorithm is used to eliminate the inverse operation. Then some experiments are implemented to verify the effectiveness of the proposed improved value function approximation algorithm. It is proved to have improved the generalization performance of reinforcement learning in continuous space.