抄録
An obstacle position estimation method using Support Vector Machine (SVM) is proposed for multi-legged robot. SVM is one of machine learning method. The proposed method evaluates a servo error on each support leg, when an obstacle contacts with the robot body. For effective data learning, efficient data acquisition method should be implemented. Since the mobility of the tip of leg depends on the manipulability ellipsoid, virtual learning data sets are spread in the ellipsoid for the effective learning. The proposed method is implemented to a real multi-leg robot, and the experimental results show the effectiveness of the proposed method.
