An Application of Vision-Based Learning for a Real Robot in RoboCup - A Goal Keeping Behavior for a Robot with an Omnidirectional Vision and an Embedded Servoing

We have applied Q-learning method so that a mobile robot acquire a shooting behavior of a soccer game. Thought the acquired behavior is tested in RoboCup-97, the performance is limited because of the narrow visual angle of the robot. Therefore, we install an omnidirectional vision into a mobile robot to enlarge a visual angle and apply Q-learning to behavior acquisition of the robot. An attention control method is also proposed for an omnidirectional vision system by means of an active zoom mechanism to reduce the learning time. We select a a goal defending behavior as a task and perform computer simulation and real robot experiments to show the validity of the proposed system.