Learning-Based Fault-Tolerant Optimal Control for Spacecraft Constrained Reorientation

A fault-tolerant approximate optimal control law for constrained spacecraft attitude reorientation is proposed in this article, where both pointing and angular velocity constraints are taken into consideration. A judiciously designed term and two barrier functions are introduced into the reward function to cope with actuator faults and double-level state constraints. Then, adaptive dynamic programming approach is employed to solve the optimal problem online, wherein a single-critic neural network (NN) is developed within the framework of reinforcement learning. Specially, the concurrent learning method is incorporated into the NN update law to relax the excitation condition for error convergence. We prove in the Lyapunov sense that, the derived approximate optimal control policy can guarantee the boundedness of state and NN estimation error while satisfying pointing and angular velocity constraints, despite the presence of actuator faults. Numerical simulations are also presented to demonstrate our control scheme.