This research examines the learning of cooperation within teams at the same time as competition between teams, using physical robots. Basic behaviors are given to the robots, which will learn to select the proper actions for given situations. I am interested in the strategies that will develop, and of course the trade-off between individual and team reward.
The reward function for learning will be based on position of the ball relative to the team's goal, personal possession of the ball, team possession of the ball, opponent possession of the ball, and goals. Perhaps some for blocking and/or stealing.
The behaviors, conditions, and messages that seem necessary to the task are listed in the illustration above.