Abstract
This paper proposes a method to associate a robot's actions and their effects of the nature of contingent physical event with force sensors, a microphone and a camera. We verified the method with a humanoid robot. The robot gets environmental information through contingent physical events. In this study where the robot attempts grasping a cylinder by hand, it gets two dimensional positions by vision processing, sounds by dropping them down and movement data by finger joint angle. These sensor data are transformed into state space by PCA processing. If a PCA data group is distributed near the centroid which is found by k-means clustering, the data are considered to be from the same sensor.
