Development of the experimental system that can acquire the gait data online in a quadruped robot

Gait patterns in animals has been investigated widely to elucidate how locomotion is generated spontaneously. The gaits of four-legged animals involve various types of patterns, such as walk, trot, and gallop, which are changed cleverly according to the locomotion speed or environmental circumstances. In particular, the way in which animals can reproduce various manners of locomotion is still not revealed entirely and is open to discussion. To investigate the contributions of mechanical systems to gait pattern transitions, we develop a minimal quadruped robot with one degree-of-freedom legs and focus on the realization of various gaits. In addition, we construct an experimental system that can extract and integrate feature quantities of movement from sensor data online into multiple devices. The developed system enables us to obtain the latent features of movement in locomotion patterns based on machine learning. We implement an auto encoder to reduce the dimensions of these data and confirm that latent features of movement can be projected to less spaces of hidden layer in auto encoder.