Application For Decision Of Soccer Based On Reinforcement Learning

Robot soccer, a popular research topic at present, involves artificial intelligence, robotics and intelligence control, etc. The entity robot soccer has a higher require on the hardware, it severely limits the development of robot soccer game. In order to put the intelligent algorithm used in robot soccer control, it is necessary to develop the simulation platform of robot soccer. Simulation platform only need a simple equipment, it can simulate entity game through the software platform. Robot soccer's simulation has become an important branch in the soccer robot research field.This paper mainly work is put the Sarsa learning algorithm applied to the decision making of soccer robot simulation, and confirm the decision making in MSRS 11vs11 soccer robot simulation platform, and make a comparison with other geometry algorithm. Because of the environment state of the simulation is a complicated and continuous state space, we need to take the continuous state space discretization first before using Sarsa learning algorithm, the result of discretization is a critical to the success of the final study. Then we should design a good reward function and action function; Reward function affects algorithm's convergence; Action function affects the soccer robot's action directly and affects the learning effect indirect. Finally, the two geometry algorithm is applied to the simulation platform, we verify the validaty of Sarsa algorithm compared with the related experimental data of the two geometry algorithm.