Neuroevolution learning for CPG-RBFN legged locomotion

Designed an Evolutionary algorithm to optimize adaptability of model-free CPG-RBFN legged robot controller, overcoming limitations in Probability-Based Black Box Optimization and enhancing robustness to noise.

Demonstrated empirical evaluations on the Half Cheetah environment, validating the neuroevolution policy’s capability to learn walking behaviors amid noise and diverse inputs.

CPG-RBF network architecture

Above image shows the CPG-RBF network architecture

Github Repository

The repository contains the research paper and the code for the project.