Structured intelligence for cyclic learning based on spiking-neural network for human friendly robots

This paper discusses a perceptual system for intelligent robots. Robots should be able to perceive environments flexibly enough to realize intelligent behavior. We focus on a perceptual system based on the perceiving-acting cycle discussed in ecological psychology. The perceptual system we have proposed consists of a retinal model and a spiking-neural network realizing the perceiving-acting cycle concept. We apply our proposal to a robot arm with a three-dimensional (3D)-range camera. We verified the feasibility of the perceptual system based on perceptual element modules such as the contrast of depth or luminance information through table cleaning task. However, our proposal could not detect dish postured or position. In this paper, we propose another perceptual module based on 3D surfaces and verify the potency for detecting dish postured or position. As experimental results a perceptual module based on 3D surfaces is effective for detecting a dish posture or position from unsteady 3D measurement information.