Behavior-Based Vision on a 4 Legged Soccer Robot

In this paper the architecture of a 4 legged soccer robot is divided into a hierarchy of behaviors, where each behavior represents an independent sense-think-act loop. Based on this view we have implemented a behaviorbased vision system, improving performance due to object-specific image processing, behavior-specific image processing and behavior-specific self localization. The system was tested under various lighting conditions, off-line using sets of images, and on-line in real tests for a robot in the role of goalkeeper. It appeared hat the performance of the goalie doubled, that it could play under a wider range of lighting and environmental conditions and used less CPU power.