This article proposes a learning-based adaptive sliding-mode control for the underactuted deployment of tethered space robot with limited input. The reduced-order system on the sliding surface is optimized by the crit...
详细信息
This article proposes a learning-based adaptive sliding-mode control for the underactuted deployment of tethered space robot with limited input. The reduced-order system on the sliding surface is optimized by the critic neural network-based nearly optimal approximation algorithm, and both the linear and nonlinear situations are taken into account. To remove the assumption that the state-dependent disturbance is vanishing, the behavior of state synchronization on the sliding surface is revealed for the first time, and the nearly optimal solution of limited input for generating stable trajectory can be approximated by the actor neural network-based adaptive dynamic programming. The hyperbolic tangent of desired input is synthesized into the optimal value function, and it can be obtained by the actor-critic neural network-based reinforcement learning algorithm solving the corresponding Hamilton-Jacobi-Bellman equation. The numerical simulation takes the numerical example of the deployment of tethered space robot to verify the effectiveness of the proposed method.
暂无评论