EFNQL: An Efficient Method Based On Q-learning And Fuzzy Neural Networks In Robocup Soccer Simulation

Robocup Soccer Simulation is considered to be one of the world’s most prestigious and large-scale robotics events, attracting researchers from all over the world. In Robocup Soccer Simulation 2D, the challenge has always been to organize intelligent agents to collaborate and analyze scenarios in order to develop more effective strategies for launching shots and dribbling. To address this challenge, an efficient algorithmic architecture called EFNQL has been developed that utilizes fuzzy control, fuzzy neural networks, and Q-learning to optimize the strategy of the agents in Robocup Soccer Simulation competitions. The architecture first establishes fuzzy rules using fuzzy control theory, then processes more complex information on the playing field using fuzzy neural networks, and finally completes fuzzy reasoning by applying Q-learning related theory for learning. Through extensive experimentation, this thesis found that the scoring rate and ball control rate were significantly improved, ultimately resulting in a second-place finish at the 2022 RoboCup China Open.