Event

CPG-ENN: A Trainable Quadruped Locomotion Framework

1 reports · 1 sources

Xiyan Liu, Yide Liu, Guoyong Mao, and Shaoxing Qu propose CPG-ENN, a trainable quadruped locomotion framework that combines explicit neural coordination with learned locomotion implementation. The framework uses a symmetry-based eight-neuron central pattern generator (CPG) to provide gait rhythms, and a pretrained execution neural network (ENN) shared across legs maps rhythm signals and trajectory parameters to joint angle references, without requiring online inverse kinematics; a single feedback policy adjusts six physically meaningful trajectory variables. The authors report that in simulation, across all 20 directed gait pairs and all combinations where both source and target speeds are drawn from {0.5, 1.0, 1.5, 2.0} m/s, the gait switching success rate was 99.39%; experiments on a Unitree Go2 validated five gaits and their switching.