Musculoskeletal robots using McKibben-type pneumatic muscles have been known to be capable of dynamic locomotion, such as walking, running, and jumping. However, most of these robots have been controlled by trial and error, such as adjusting the timing of opening and closing of pressure valves. We have focused on the antagonistic drive, a characteristic of musculoskeletal systems. Our previous study proposed “tension feedback cooperative control,” which generates periodic motion by utilizing autonomous coordination of antagonistic muscles. In this paper, we applied this control law to a two-degree-of-freedom leg model with antagonistic bi-articular muscles that mimic the hind legs of a four-legged walking animal. The autonomous coordination of the antagonistic bi-articular muscles and the generation of periodic motion were verified by simulation and experiment on a real machine. The simulation results confirmed that the antagonistic bi-articular muscles cooperate autonomously with the proposed control law. We also confirmed that by changing the feedback gain, five types of cooperative modes are generated, and periodic motions like walking, standing still, and running of animals are generated autonomously. Furthermore, experiments on an actual machine confirmed three cooperative modes of antagonistic bi-articular muscles. And periodic motions like walking, standing still, and running were generated autonomously.
