Behavior Learning for a Mobile Robot with Omnidirectional Vision Enhanced by an Active Zoom Mechanism

摘要

The authors propose an attention control method for an omnidirectional vision by an active zoom mechanism. It is implemented by controlling focal length of the camera without pan or tilt mechanism. We install an omnidirectional vision with a hyperbolic mirror to a mobile robot and apply Q-learning for its behavior acquisition. In a goal defending behavior of a soccer game, the robot can learning the behavior and the attention control works to reduce learning time. In this paper, we explain our method and experiment. 1 Introduction Omnidirectional vision has been receiving increased attentions as a method to capture the whole view all around the imaging system at any instant in time. This causes a wide variety of applications which include autonomous navigation [1], visual surveillance and guidance [2], video conference, virtual reality, and site modeling [3]. Since the omnidirectional vision can capture the visual information from the all directions, no pan or tilt control is necessar...