Ensuring UAV safety: A deep learning-based fault-tolerant approach with control barrier functions

Accurate control of unmanned aerial vehicles (UAVs) is vital for precise navigation and reliable execution of tasks, particularly in safety-critical scenarios. Nevertheless, actuator faults, especially rotor failures, can significantly impair UAV performance and compromise safety. This paper proposes a robust control architecture that integrates deep learning with finite-time control methods, further augmented with control barrier functions to enforce safety constraints rigorously under unknown actuator faults. A deep recurrent neural network (RNN) estimator is incorporated to identify and compensate for actuator faults in real-time. At the same time, a finite-time control mechanism ensures rapid and precise trajectory tracking. Moreover, including control barrier functions is an additional protective layer, robustly guaranteeing that the UAV remains within designated safety boundaries despite uncertainties, disturbances, and severe actuator faults. We validate the effectiveness and practicality of the proposed method using a modified five-rotor UAV model tailored for heavy-load transport tasks. Detailed finite-time stability analysis and comprehensive simulations demonstrate the practicality and enhanced performance of our proposed approach. Simulation results demonstrate superior accuracy in trajectory tracking and robust enforcement of critical safety constraints, such as maintaining a safe altitude threshold in hazardous conditions. These findings underscore the potential of the proposed framework for reliable and safe UAV operations in demanding environments.