Simultaneous Learning and Recalling System for Wholebody Motion of a Humanoid with Soft Sensor Flesh

Abstract

In order for robots to be taught new motions by human beings in any time they are directly touched, seamless ongoing learning system is necessary. In this paper, we focus on constructing ongoing learning system with simultaneous learning and recalling function for acquiring whole-body posture. Here, learning state monitor that checks a desirable state for learning and controlling when to learn continuously, a fast neural network model dealing with temporal correlative relationships, and a humanoid with soft sensor flesh that can enable itself to detect the touched place and to adapt to the humans, are key issues. Furthermore, proposed system is confirmed by a simultaneous learning and recalling experiment using an actual humanoid with soft sensor flesh.