Self-learning control of cooperative motion for a humanoid robot

Abstract

This paper deals with the problem of self-learning cooperative motion control for a heavy work of a humanoid robot in the sagittal plane. A model with 27 linked rigid bodies is developed to simulate the system dynamics. A simple genetic algorithm (SGA) is used to find the necessary torques in each joint to obtain a desired cooperative motion, which is to minimize the total energy consumption, for the humanoid robot's postures of trunk and hands. And the multilayer neural network using the backpropagation is also described in order to control the system in real time