Reliable trajectory tracking for quadrotor UAVS under uncertainty: a sliding mode integral reinforcement learning approach

To address the challenge of achieving precise trajectory tracking for quadrotor UAVs operating in uncertain dynamic environments, a control strategy based on Sliding Mode Integral Reinforcement Learning (SIRL) is proposed. This approach combines the high robustness of sliding mode control with the adaptive online optimization capability of integral reinforcement learning. To improve learning stability, a Target Network mechanism is incorporated into the Critic network of the SIRL framework. By providing a consistent learning reference, this mechanism effectively reduces oscillations in value function estimation and facilitates the smooth convergence of the learned policy. A cascade control structure is implemented, wherein the outer loop utilizes the SIRL controller to derive an optimal policy for generating desired attitude commands, thereby enabling accurate position tracking. Simultaneously, the inner loop employs an Incremental Nonlinear Dynamic Inversion controller to achieve rapid and stable control of attitude and altitude. The proposed controller is evaluated through simulation across a range of complex trajectories, including circular, helical, and figure-eight paths. The results indicate that the controller effectively learns to compensate for system uncertainties and demonstrates significant advantages in tracking precision, robustness, and input constraint management. The incorporation of the target network plays a key role in ensuring stable learning performance.