Braitenberg Soccer

Well-developed individual and collaborative skills, such as dribbling the ball, positioning, and passing are required for a team of robots to be successful against an opponent team in a robot soccer scenario. This paper proposes an approach to individual and collaborative skill learning, where the robots are modeled as Braitenberg vehicles, and the required skills are implemented as combinations of very primitive behaviors. Without explicit communication and role assignment mechanisms, the robots were able to learn how to play soccer as a team after a short training session, where reinforcement learning was used to construct the optimal state-action mapping. Experiments demonstrate that a team of robots can indeed learn to play soccer reasonably well without using complex environment models and state representations.