Analysis of a neural-network-based adaptive controller for deep-space formation flying

Abstract The high-precision control problem of deep-space formation flying is considered in this paper. To achieve accurate position control, a neural-network-based adaptive controller is adopted as a benchmark to counteract the unknown perturbations in deep space. This controller is enhanced by considering the dynamic environment of deep-space formation, yielding a proportional-integral-derivative (PID) adaptive controller. Compared with the benchmark, the PID adaptive controller has a simpler architecture with fewer parameters. Therefore, the PID adaptive controller avoids the challenge of parameter tuning, which is a longstanding disadvantage of neural-network-based adaptive controllers. Simulations show that the adopted neural-network-based adaptive controller and the PID adaptive controller have nearly equivalent adaptive capabilities.