Experimental evaluation of adaptive neurocontrollers for a prototype space robotic manipulator

Abstract

The dynamics of robotic systems are often affected by poorly modeled internal and external forces. This paper presents an experimental evaluation of adaptive compensation for unmodeled dynamic effects on a prototype space robotic manipulator. Recent extensions to adaptive nonlinear control theory show that neural networks can be used to reliably compensate for these effects. In this paper, a newly developed neurocontrol algorithm is implemented on a prototype robotic manipulator with a single degree of freedom. The experimental results demonstrate that by adaptively compensating for friction torques in the motor joint and for hydrodynamic effects present during underwater operation of the arm, the neurocontroller greatly improves the tracking performance of the system, reducing tracking errors observed witli a simple PD controller by an order of magnitude. Since the neurocontroller is capable of compensating for unmodeled forces, it seems well suited for space robotic operations which occur in unfamiliar surroundings with poorly modeled environmental factors, as well as for neutral buoyancy simulation of space operations.