From Biological Gait to CPG Network Design: A Biomechanics-Driven Model-Selection Pipeline

Transferring observed gait coordination into executable Central Pattern Generator (CPG) specifications remains challenging because networks are often tuned empirically without systematic links to measured biomechanics. This study presents a biomechanics-driven pipeline using the walking gait of the fiddler crab Uca pugnax. Marker-based video analysis provided mean dominant frequency, movement amplitudes, phase relations, and coordination hierarchy. These descriptors defined oscillator roles, reference hierarchy, coupling topology, phase targets, communication direction, and an eight-oscillator architecture. A coarse-to-fine procedure selected parameters for phase-only Kuramoto and limit-cycle Hopf networks under unidirectional and bidirectional communication. Offline simulations assessed amplitude, phase-offset, convergence, dominant-frequency, and residual-oscillation errors. Kuramoto converged in 0.098 s under both communication schemes, with mean phase errors of 0.40% and 0.08%, respectively; both communication configurations preserved the prescribed gait frequency, with the bidirectional network producing 0.522 Hz. Hopf preserved frequency within 0.02% and 0.01%; the selected unidirectional configuration achieved the lower phase error, whereas the selected bidirectional configuration yielded a lower mean amplitude error (0.32% versus 2.87%), with longer convergence and residual oscillations. The pipeline provides a reproducible route from biological measurements to computational CPG design and model selection before physical deployment; robot validation is outside the present scope.