Abstract
An anthropomorphic soft finger with many tactile receptors is supposed to have high tactile sensing ability. However, it is more difficult to control a robot equipped with the fingertip because it is too difficult for the designer to calibrate all the receptors randomly distributed in it. One of the approaches to realize dexterous manipulation by using it is skill acquisition through a task. In this paper, we propose a learning mechanism to acquire a skill for Peg-in-Hole task by using tactile sensor and vision sensor. We show a preliminary experimental result to show the effectiveness of the proposed method.
