A Simple Rule for Quadrupedal Gait Transition Proposed by a Simulated Muscle-driven Quadruped Model with Two-level CPGs

Abstract

This study aims to figure out the principles of gait generation and transition in quadrupeds by using a simulated quadruped model. We design a neuromorphic controller to drive the realistic muscle-driven legs of a cat model. Each leg has three joints driven by six muscle models, including two-joint muscles. We apply a two-level central pattern generator (CPG) to control each leg, which synergistically drives different sets of the muscles in each of the four phases (i.e., stance, lift-off, swing, and touch-down) and which can reproduce catlike leg trajectories. Consequently, we enable the simulated cat model to safely locomote in three-dimensional space. Although the CPGs inherently generate a trot pattern as a default gait, because of mutual inhibitory connections between the lateral neighboring CPGs, the gait autonomously transitions from walking to trotting to galloping, according to the speed. Note that the walking and galloping are not preprogrammed. This is caused by leg loading feedback to the CPGs, which was installed to maintain balance. Therefore, we hypothesize that each gait is generated by the differences in posture that arise at different speeds, and that there is a relationship between gait and posture control.