Enhancing Decisions of Goalkeeper and Kicker Players in the RoboCup 2D Simulation League through Behavioral Cloning

RoboCup is recognized as a landmark project in the robotics and artificial intelligence industry, providing a crucial platform for the advancement of research and innovations in autonomous and multi-agent systems. This study delves into the use of behavioral cloning in the RoboCup 2D Simulation League to imitate the best decisions and improve the performance of the goalkeeper along with the kicker players. To improve the performance of our soccer team in the league, we have developed a new approach that combines behavioral cloning and game log parsing. To achieve this, we played extensively against the champion team Helios and analyzed the actions and decisions of their goalkeeper and kicker players. This data was used to create a large dataset that our team could use to improve their imitation skills. Our innovative log generator tool and game logs parser made the labeling process easier and will also be useful for future research in the league. Novel features are extracted to represent the state of the game considering the limitations of each player including their head direction along with their field of view. Next, a neural structure is trained to imitate the behavior of kicker players and the goalkeeper. Moreover, the time limit for decision-making in each turn is considered. This study showed that using behavioral cloning for imitation learning noticeably improved our team’s actions and decisions due to the enhancement of the number of accurate passes, on-target shoots, and saves by the goalkeeper.