Benchmarking Open-Loop Central Pattern Generator Controllers for Lower-Limb Exoskeleton Walking Trajectories

Central Pattern Generators (CPGs) enable rhythmic, biologically inspired control of robotic locomotion. This study presents a comparative evaluation of three CPG models: Hopf, Matsuoka, and sinusoidal implemented on a simplified lower-limb exoskeleton with three actuated joints per leg (hip, knee, and ankle). Each controller was simulated in MATLAB and validated using IMU measurements from a physical prototype. All models reproduced the fundamental gait frequency (~0.6 Hz), though accuracy varied: hip RMSEs were below 1°, while knee and ankle errors reached up to 19° for Hopf. PCA analysis revealed that Hopf generated clean and stable cyclic patterns, the sinusoidal method produced smooth but less adaptive motion, and Matsuoka introduced richer variability with reduced regularity. Power analysis under loaded and unloaded conditions showed similar average consumption (~4.9 W) but distinct stability profiles. These results demonstrate the trade-offs between simplicity, adaptability, and energy robustness in selecting CPG controllers for exoskeleton locomotion.