Abstract
In this paper, we propose a method for predicting result of lifting-up an object by using the data collected in advance. In the case of actual robot system, it is difficult to collect a large amount of data, and hence we propose a method for efficient data collection based on the observation data. In this method, tactile information is acquired while lifting-up an object with humanoid robot to collect date in advance. Using the senser information, the robot estimates a posture that will be effective for predicting result of lifting-up by an analysis on the collected data after collecting a certain amount of data. Predicton of lifting-up is based on canonical correlation analysis with tactiles senser information and image information which is acquired after the lifting-up motion. The proposed method was verified by experiment using humanoid robot equipped with tactile sensors.
