Application of Reinforcement Learning in Half Field Offense of Robot Soccer

This article main introduce the applicatiion of reinforcement learning in half field offense to robot soccer, the environment of robot soccer is a continuous state space, we should discretize the state of environment,we define the state using a set of variables. In order to overcome the shortcoming of the agent update Q value independent, we adopt communication between robots to update Q value of all offense agent.Finally we perform an experiment in 4V5 half field offense, and get an ideal result.