Event/self-triggered cooperative control via reinforcement learning for a quadrotor team under multiple faults and denial-of-service attacks

The cooperative fault-tolerant control problem of a quadrotor team, subject to multiple faults and denial-of-service (DoS) attacks, is addressed via event/self-triggered strategies and reinforcement learning. Multiple under-actuated quadrotors with nonlinearities, couplings, and unknown dynamical parameters are introduced to achieve distributed cooperation. Event-triggered observers are developed to estimate position references under uncertain cyber faults and DoS attacks. Observer-based optimal controllers for the nominal position and attitude subsystems are learned by off-policy reinforcement learning without dynamical information. Fault-tolerant controllers are constructed by integrating the learned controllers and adaptive actuator fault compensators. A self-triggered strategy is further proposed to avoid continuous communication. The relationship among the time-varying topology coupled with cyber faults, the features of DoS attacks, and the closed-loop control system stability is analyzed, and the exclusion of Zeno behavior is proved. Simulation results illustrate the effectiveness of the proposed methods.