Experiments Toward On-Line Identification and Control of a Very Flexible One-Link Manipulator

Abstract

An adaptive control algorithm based on the self-tuning regu lator concept has been experimentally investigated for appli cation to a very flexible one-link robotic manipulator. Adap tive control is an attractive methodology for maintaining the performance of precise controllers designed for such manipu lators under conditions of varying end-effector load. The use of noncollocated sensors and actuators to give good accuracy in tip positioning also places stringent requirements on the accuracy of dynamic models used for controller design. Identification and control design techniques suitable for on-line implementation have been demonstrated experimen tally on such a single-link flexible manipulator. The identifi cation algorithm employed is a filtered version of the recur sive least-squares algorithm. It is a development of algorithms previously used with the Stanford four-disk sys tem. Stable controllers with good step responses were de signed using the system models identified with the identifica tion algorithm. The necessity of filtering sensor data to achieve accurate identification was motivated analytically and confirmed experimentally. Accurate identification of 2 system transfer functions was achieved with a 4-s-long data record. The algorithms demonstrated could be used in an adaptive-learning setting to improve the performance of a ro botic system subject to varying loads.