Learning-based online optimal sliding-mode control for space circumnavigation missions with input constraints and mismatched uncertainties

Abstract

A learning-based online optimal sliding-mode control strategy is developed for space circumnavigation missions subject to input constraints, and the mismatched uncertainties caused by measurement uncertainties are also considered in this infinite-horizon optimal control problem. The logarithmic hyperbolic cosine function is used to design the optimal value function to overcome the weakness that the derivative of the adaptive weight of neural network (NN) changes too fast when the value of the sliding-mode function is too large, and another suitable nonquadratic function is used to incorporate input constraints into the optimal control framework. To approximate the Hamiton-Jacobi-Bellman equation corresponding to the novel optimal value function, an actor-critic (AC) architecture is introduced with NNs, and a finite-time disturbance observer (FTDO) is employed to estimate the mismatched uncertainties in the plant. The simulation results verify the effectiveness of the proposed approach.