Obstacle Avoidance for AUV by Q-Learning based Guidance Vector Field

Autonomous obstacle avoidance technology is the key to determine whether the autonomous underwater vehicle (AUV) can reach its destination safely. The Q-Learning based guidance vector field is proposed to solve the 3-dimensional obstacle avoidance problem for AUV. Firstly the initial guidance vector field in free space is constructed to guide AUV reach the destination along the shortest path. Then the modulation matrix is proposed to quantify the influence generated by the obstacles so that the modified guidance vector field in obstacle environment is obtained. For the case of AUV entering the trap area, the Q-learning algorithm is used to find the shortest path from the current position to the destination. A virtual target is chosen to guide AUV escape from the trap area. Finally the simulation results show that this method is useful for AUV to complete the obstacle avoidance task in complex marine environment.