Neural networks for force control of an assembly robot

摘要

The performing of robotized assembly tasks implies hard constraints like the knowledge of robot dynamics and robot-environment interaction. So classical form control laws lead to low performances. The aim of this paper is to show the ability of neural techniques for learning control under real world conditions of unknown non-linear systems. In our control strategy neural networks are used for the identification and control of an assembly parallel robot performing a peg-in-hole insertion. Such a system is strongly non linear. In this paper, we propose a new approach of hybrid force position control based on feedforward neural networks. The proposed structure does not need any mathematical modelling of system dynamics. Experimental results show that identification and control schemes suggested in this paper are practically feasible.