Collision avoidance controller for AUV systems using stochastic real value reinforcement learning method

Based on the basic principles of the reinforcement learning and also motion characteristic of an AUV system, named Twin Burger 2, a collision avoidance algorithm is proposed here. Most of the researches in reinforcement learning have been done on the problems with discrete action spaces. However, many control problems require the application of continuous control signals. In this research we are going to present a stochastic real value reinforcement learning algorithm for learning functions with continuous outputs. Obstacle avoidance mission is divided into targeting and avoiding behavior. Because of the complexity of the implemented method, only targeting results, which are achieved most recently, are proposed here and research is under progress to achieve to the final goal.