Research on a Multigent Reinforcement Learning Algorithm for Warship Formation Transformation

Abstract

The current study of multiagent reinforcement learning has been described and an improved method which can make agent use the mixed Q_learning and neural network has been introduced to form the team and avoid collision between each other members in a warship fleet under multiagent environment. In the multiagent algorithm, many problems as difficult information exchanges between agents, an extraordinary searching space and low robust of the system are studied and a new method is promoted to deal with these problems very well. At last, an example is given to illustrate the validation of the reinforcement learning algorithm.