Design of an Effective Trajectory Control Method for Quadruped Robot via Neural Network

Abstract

In this paper, a novel trajectory control method based on ensemble neural networks for quadruped robot is designed. A proportion and differentiation (PD) neural network, which includes foot-trajectory control unit (FTCU) and steering control unit (SCU), is constructed to control the motion trajectory of quadruped robot. After that an advanced particle swarm optimization (PSO) is introduced to optimize the weights of PD neural network. In addition, the FTCU based on RBF neural network (RBF-NN) and SCU based on general mathematical model are cooperated to realize a new trotting gait. The compared results of simulations show that the FTCU and SCU can be used to realize the gait and steering control of quadruped robot accurately. Furthermore, in comparison with existing methods, the proposed control system has excellent accuracy and obvious advantages on anti-interferences, either for inner disturbances of system (IDS) or for exterior stochastic disturbances (ESD) caused by irregular terrain.