Evolution of central pattern generators for the control of a five-link bipedal walking mechanism

With the aim of producing a stable human-like bipedal gait, a five-link planar walking mechanism is coupled with a central pattern generator (CPG) neural network, consisting of units based on Matsuoka's half-center oscillator model with a firm basis in neurophysiology. As a minimalistic approach to bipedal walking, this type of walking mechanism contains only four actuators, and is lacking feet and ankles. The mechanism is simulated with accurate physics, allowing realistic fitness evaluations for the creation of CPG controllers through evolutionary computation. The oscillatory parameters, internal connectivity structure, and external feedback pathways of the networks are determined through genetic algorithms (GA) optimization. The evolved CPG networks are transferred to a hardware implementation of the mechanism, to test their performance under real-world dynamics. Results confirm that the biologically inspired CPG model is very well suited for controlling legged locomotion, since a diverse manifestation of CPG networks (with and without external feedback) have been observed to succeed during the course of GA evaluations. Observations also imply that while the CPG mechanism is inherently able to sustain a stable gait, the utilization of feedback pathways makes the gait more human-like and is needed to provide a means to adapt to irregularities in the environment.