Abstract
A learning impedance control approach based on neural networks is presented for a robot to accomplish some contact tasks. Firstly, a velocity-based discrete-time impedance control law is obtained to control the position and contact force of the robot in the same direction. Secondly, a computational method for generating the reference inputs of the robot control system is given for the contact task of the robot. Thirdly, on-line learning algorithms using neural networks are developed to adjust the inertia, damping and stiffness parameters of the robot in the unknown contact environments. Lastly, the effectiveness of the present approach is verified by pressing a spring using a 6 degrees of freedom robot. The adaptiveness, stability and flexibility of the present approach are also confirmed.
