Acquisition of Motion Planning Algorithm of Autonomous Underwater Vehicle Considering Time Cost by Reinforcement Learning and Teaching Method

A training algorithm for motion planning of non-Holonomic Autonomous Underwater Vehicle ( AUV ) by reinforcement learning is proposed in this papaer. The acquired motion planning algortihm can be applied to an environment with arbitrary configuration of obstacles and the strong water current. The AUV can appropriately reach to the destination point in severe environmental condition by the algorithm and the operator of the AUV can designate arrival time cost. In order to avoid state space explosion and make up for undesirable non-Markovian Effects which comes from the characteristics of the dynamics of AUV, hierarchical reinforcement learning algorithm is introduced. Results of the simulation show high learning speed and motion planning ability of the acquired motion planning algorithm in severe disturbance.