連想記憶に基づた多指脚ロボットの歩行動作獲得

抄録

For autonomous hexapedal locomotion of a legged robot, SHIKYAKU, we developed a teaching playback system. Since the mechanism of SHIKYAKU is based on the extension and flexion of fingers, motion patterns of the robot for walking and climbing are taught with finger through LeapMotion. The robot records the motion patterns as attractors. At the same time, the robot also records the sensor data. Thus the robot is allowed to have the motion attractors corresponding to the environment. In the teaching playback phase for the autonomous hexapedal locomotion, we use two feedforward neural network sequentially. The first one is for identifying the environment from the senser inputs. The second one is so-colded an associative memory for deciding locomotive motions. Through simulation experiments, we show that the robot is enabled to walk on the flat plane and climb an obstacle autonomously.