Reinforcement Learning-Based Optimized Tracking Control for Uncertain Robot Manipulators with Composite Constraints

摘要

To enhance the safety performance of robotic manipulators, this study proposes a control strategy that integrates optimized backstepping technique with composite constraints. Initially, to address the problem of symmetric state constraints, a barrier Lyapunov function that combines with the prescribed performance functions is introduced, which provides safe boundaries for the operation of the system. Meanwhile, a compensation method is developed to effectively eliminate the impacts of input saturation. Subsequently, an actor-criticidentifier architecture is established within the reinforcement learning, where optimal control policies are realized through iterative learning. Neural network approxim-ation methods are used to accurately estimate the uncertainty of the system while maintaining the convergence of the learning process. Finally, all signals in the closed-loop system are proven to be bounded via Lyapunov-based stability analysis. Simulations are provided to verify the effectiveness of the proposed strategy.