Adaptive Robust Attitude Control for Multiple Quadrotor Systems via Integral Reinforcement Learning

This article explores the adaptive robust optimal attitude control issue of multiple quadrotor systems with unknown external disturbances and actuator faults. To achieve the desired tracking performance, we introduce optimal control theory to improve control accuracy and energy utilization efficiency. Considering the unavailability of disturbance information and the bias signals caused by actuator failures in the system model, we design an integral reinforcement learning (IRL) control algorithm that eliminates the requirement for system dynamics. Unlike more complex structures such as identify-actor-critic and actor-critic, we employ a critic-only neural network (NN) to evaluate system performance and execute control actions, thereby significantly reducing the computational burden. By combining experience replay (ER) and gradient descent techniques, we derive the update laws of NN weights, eliminating the need for persistent excitation (PE) condition. Finally, the simulation experiments demonstrated the practicability of this control strategy.