Robot contact motion control using environment impedance identification: Application of neural networks.

Abstract

The robot manipulators must have capability of controlling mechanical interaction with objects which are involved in various tasks. In order to control the mechanical interaction, the environment impedance which is not always known has to be identified. In this thesis, environment impedance on-line identification methods based on Recursive Least Squares Estimation and Neural Networks approach are investigated in order to achieve an efficient force control. Experiments were done for both methods with the 3-DOF Direct-Drive planer robot manipulator, by applying a force on the environment with a robot manipulator and measuring the force at the same time with a force sensor. The hybrid controller is used for controlling the robot for environment impedance identification. The nonlinear three-layer neural networks feedforward direct controller using a dynamic back-propagation method based on past input/output information of the plant is also proposed. This controller is used for a 3-DOF robot applying force to the environment. (Abstract shortened by UMI.)