Abstract: 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 simplifi ed rotary joint. In addition, Q-learning, a type of machine learning, was used to fi nd the optimal motion profi le of the spinal joint. The bounding gait was implemented on the robot system using the motion profi le of the spinal joint, and it was confi rmed that using the spinal joint can achieve a faster Center of Mass (CoM) forward speed than not using the spinal joint. Although the motion profi le 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 profi le, which is sinusoidal, was similar. 

Key words: Quadruped robot system, Bioinspired methods, Legged robots, Machine learning, Q-learning