Abstract
This paper describes a gait generating network for a hexapod robot under various curvature turnings. Formica japonica Motschulsky, that is one of major ant with about 6 [mm] body length in Japan, was used for a gait analysis. Movements of the leg joints during free walking was recoded at 60 to 250 [frame/s] by a high-speed video camera. From the data, we found that the turning gaits were classified into two distinct patterns: a sharp turning gait in which case the radius of curvature is under 5 [mm], a gradual turning gait in which case that is 5 to 40 [mm] . A turning gait generating network, which was possible to generate each gait and to vary a radius of curvature, was devised. That is a neural network which is described by neurons and connections between them using excitation and inhibition. We made a hexapod robot with 3 DOF each leg and verified the accountability of the gait generating network using the robot. As a result, smooth and stable turning in different curvatures was achieved.
