Dynamic Walking Control of Quadruped Locomotion Robot Using Neural Network.

Abstract

This paper proposes a new neural network algorithm for dynamic walking control of quadruped locomotion robot. A two-degree-of-freedom (2 DOF) control system, such as that in a locomotion robot, needs feedforward control in addition to feedback control to improve the control performance in terms of transient tracking responses. We believe that most research work on quadruped locomotion robots uses local feedback loops with PD control. We propose a new control strategy for a quadruped locomotion robot with dynamic locomotion using the 2 DOF control system with adaptive learning based on feedback error and learning based on stratum. Also, we propose an algorithm for compensation of a reference trajectory by the improved neural network. This system consists of a decentralized controller and a centralized controller. We have succeeded in conducting an actual locomotion test with very smooth and fast trotting locomotion.