Biomimetic Quadruped Robot with a Spinal Joint and Optimal Spinal Motion via Reinforcement Learning

Feline animals can run quickly using spinal joints as well as the joints that make up their four legs. This paper describes the development of a quadruped robot including a spinal joint that biomimics feline animals. The developed robot platform consists of four legs with a double 4-bar linkage type and one simplified rotary joint. In addition, Q-learning, a type of machine learning, was used to find the optimal motion profile of the spinal joint. The bounding gait was implemented on the robot system using the motion profile of the spinal joint, and it was confirmed that using the spinal joint can achieve a faster Center of Mass (CoM) forward speed than not using the spinal joint. Although the motion profile obtained through Q-learning did not exactly match the spinal angle of a feline animal, which is more multiarticular than that of the developed robot, the tendency of the actual feline animal spinal motion profile, which is sinusoidal, was similar.