Predefined-Time Reinforcement Learning for Multi-AAV System Formation Control Under DoS Attack

This article studies the secure formation problem for multiple autonomous aerial vehicles (AAVs) subject to the denial-of-service (DoS) attack. An estimation-control hierarchical architecture is introduced to design the secure controllers. In the estimation layer, the resilient observers are designed to estimate the leader's states under DoS attack. In the control layer, by the time base generator (TBG) technique, the optimized predefined-time backstepping controllers are devised for the position and attitude loops. Obtaining the optimal control policies is related to solving the Hamilton-Jacobi-Bellman equations. Utilizing the critic-only neural network structure, a TBG-based predefined-time reinforcement learning algorithm is proposed to learn the optimized control policies online. The corresponding predefined-time convergence and the closed-loop stability are analyzed. Finally, simulation examples are conducted to verify the effectiveness of the presented scheme.