Reconfigurable control of a free-flying space robot using neural networks

An experimental demonstration of a new, reconfigurable neural-network-based adaptive control system is presented. The system under control is a laboratory model of a free-flying space robot whose position and attitude are controlled using eight on-off air thrusters. The neural-network controller adapts in real-time to account for multiple destabilizing thruster failures. The adaptive control system is similar in structure to a conventional indirect adaptive control system. While a system identification process builds a model of the robot, a neural controller is trained concurrently using backpropagation to optimize performance using this model. The active controller is updated every few seconds, yielding quick adaptation. Stability is restored within 4 seconds, system identification is complete within 48 seconds, and near-optimal performance is achieved within 2 minutes.