Adaptive control of free-floating space robots using "neural" networks

Abstract

With suitable modifications, almost any control methodology for fixed-base manipulators can be used to develop control algorithms for manipulators mounted on spacecraft. However, a significant exception occurs for free-floating manipulators when there is uncertainty on any of the mass properties of the system. Although, structurally, tracking controllers for these systems can be implemented using only measurements of the manipulator state, the standard adaptive techniques by which such controllers can learn to eliminate the effects of physical uncertainty cannot be employed. The linear parameterization of dynamic uncertainty exploited by standard fixed-base adaptive controllers cannot be obtained when the base is free-floating. By instead utilizing new results in multivariable adaptive nonlinear control, this paper demonstrates a controller architecture which does not require such a parameterization and hence can be directly utilized for free-floating manipulators. These new algorithms employ "neural" networks to stably "patch together" the required control input from a collection of extremely simple computing elements. The performance of the proposed algorithm is demonstrated on a simulation of a prototype NASA flight vehicle.