Ground contact detection of robot legs by discriminating angular velocities and motor current

We investigated the contact detection method of legs using an angular velocity (AV) sensor and a current sensor. When lifted legs step on an obstacle, the AV of the robot body shows different patterns from the one of normal walking. Discriminating these unusual patterns from normal one, the robot control system is able to detect stepping on an obstacle. We experimented stepping on an obstacle and recorded the change of AVs. A three layered neural network was used for discrimination. As a result, a detection rate of 96 [%] was obtained. Referring the results of detection, the incline of robot body was suppressed to 1.0 [degree] by stopping the leg when it touched an obstacle. We also measured the motor current when a free leg stepped on an obstacle. The incline of robot body was suppressed to 0.7 [degree] or less at the detection rate of 100 [%].