Learning plastic matching of robot dynamics in closed-loop central pattern generators

Abstract Animals achieve agile locomotion performance with reduced control effort and energy efficiency by leveraging compliance in their muscles tendons. However, it is not known how biological controllers learn to leverage the intelligence embodied leg mechanics. Here we present a framework match patterns mechanics based on concept of short-term elasticity long-term plasticity. Inspired animals, design robot, Morti, passive elastic legs. The quadruped robot Morti controlled bioinspired closed-loop central pattern generator that designed elastically mitigate perturbations using sparse contact feedback. By minimizing amount corrective feedback long term, learns controller its walk within 1 h. advantages mechanics, improves 42% without explicit minimization cost function.