Teaching and learning of deburring robots using neural networks

Abstract

A method for applying advanced robot adaptive control to manufacturing processes is described. A teaching method for constructing a sensor-based, task-level adaptive control system is described. Adaptive control laws that elucidate human motions are identified and stored in a multi-layer neural network. The resultant task performance is evaluated, and the relationship between the human actions and the performance index is stored in a second neural network. Based on the initial teaching data, the robot begins to perform a task. While performing a task repeatedly, the robot acquires additional data and improves its performance. Errors with respect to the performance index are propagated through the second network to modify the adaptive control law represented by its performance and excel the human operator who has provided the initial teaching data. A proof-of-concept demonstration and simulation are presented.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>