Dual learning for perception and behavior of mobile robots

Abstract

This paper deals with a learning method for perception and behavior of mobile robots. An intelligent robot should acquire its behaviors based on the embodiment of the robot itself. The robot cannot understand the meaning of the perceptual information until acquiring behaviors, in other words, the robot does not know which information in the facing environment is important for the target behavior beforehand. This indicates that the robot should reconfigure the perceptual mechanism after acquiring behaviors. Therefore, this paper proposes a coupled learning mechanism of the perception and the behavior. Furthermore, we discuss a reconfiguring method of the perceptual mechanism to cohere the relationship with the facing environment. We show the effectiveness through computer simulation results.