Terrain-adaptive locomotion is an important issue for the autonomous walking robot to reach the destination within limited time and energy. To percept the ground condition without image sensing and processing that consume a lot of energy, we investigated the artificial proprioception based on acceleration and the walking gait switching mechanism for an amoeba-inspired autonomous four-legged walking robot. We implemented the artificial proprioception mechanism by combining the 3-axis acceleration and a reservoir computing (RC) classifier. The RC model was trained offline using the data obtained from the robot walking on flat and rough grounds and the model was deployed on a commercially available microcontroller. We examined the ground condition classification in the four-legged walking robot with the trained model. We also demonstrated the switching of the walking gait in the robot between the preprogrammed normal walking and the amoeba-inspired successive searching for leg movement, depending on the ground condition. The gait switching based on our artificial proprioception was found to improve the locomotion efficiency of the robot walking on the complex ground conditions.
Yamaguchi et al. (Thu,) studied this question.