Learning-Based Passive Fault-Tolerant Control of a Quadrotor with Rotor Failure (2503.02649v1)

Abstract: This paper proposes a learning-based passive fault-tolerant control (PFTC) method for quadrotor capable of handling arbitrary single-rotor failures, including conditions ranging from fault-free to complete rotor failure, without requiring any rotor fault information or controller switching. Unlike existing methods that treat rotor faults as disturbances and rely on a single controller for multiple fault scenarios, our approach introduces a novel Selector-Controller network structure. This architecture integrates fault detection module and the controller into a unified policy network, effectively combining the adaptability to multiple fault scenarios of PFTC with the superior control performance of active fault-tolerant control (AFTC). To optimize performance, the policy network is trained using a hybrid framework that synergizes reinforcement learning (RL), behavior cloning (BC), and supervised learning with fault information. Extensive simulations and real-world experiments validate the proposed method, demonstrating significant improvements in fault response speed and position tracking performance compared to state-of-the-art PFTC and AFTC approaches.

• The paper introduces a Selector-Controller network that integrates fault detection and adaptive control into a single learning framework for quadrotor rotor failure.
• A hybrid learning approach combining reinforcement learning, behavior cloning, and supervised learning trains the policy network, leveraging expert actions and fault differentiation.
• Numerical and experimental results show superior fault response speed and position tracking accuracy compared to traditional methods, validating the approach's effectiveness in varied failure scenarios.

Learning-Based Passive Fault-Tolerant Control of a Quadrotor with Rotor Failure

This paper presents a novel approach for passive fault-tolerant control (PFTC) in quadrotors, particularly in managing arbitrary single-rotor failures without requiring rotor fault information or controller switching. As rotor failures in quadrotors typically lead to significant control challenges due to the non-redundant actuator setup, the authors propose a learning-based framework that does not switch between controllers, unlike traditional active fault-tolerant control (AFTC) approaches. This proposed methodology integrates a Selector-Controller network allowing unified policy network functionality encompassing both fault detection and adaptive control.

The research situates itself within the broader landscape of UAV safety, where rotor failure typically necessitates either AFTC or PFTC strategies. AFTC techniques often rely on an external Fault Detection and Diagnosis (FDD) module, which suffers from latency and errors in estimation, potentially degrading control performance. Traditional PFTC methods, while avoiding the complexities of fault detection, generally treat rotor failures as disturbances, leading to suboptimal control in severe fault scenarios.

This paper proposes a Selector-Controller network, which integrates fault detection with control strategies in a single learning network. The network comprises several identical controller networks tailored to specific rotor failure scenarios and a selector network determining the optimal controller to apply. This setup effectively mitigates the gradient conflicts that might arise from different rotor failure scenarios, resulting in improved learning convergence and final performance.

The policy network is developed through a hybrid framework, combining reinforcement learning (RL), behavior cloning (BC), and supervised learning with fault information. This synergistic approach not only leverages the adaptability of learning-based methods but also incorporates expert actions from AFTC scenarios to guide policy training. The integration of supervised learning ensures that the selector network can appropriately differentiate between various fault scenarios and appropriately guide controller selection.

Numerical results from both simulations and real-world experiments demonstrate the superior performance of the proposed method. The authors report enhanced fault response speed and position tracking accuracy compared to traditional PFTC and AFTC approaches, even achieving performance levels comparable to ideal AFTC methods where fault conditions are instantaneously detected. A significant numerical result is the policy's ability to maintain effective control across varied and severe fault conditions that had previously attenuated the efficacy of uniform approach methods.

The implications of this work extend beyond simple enhancements in quadrotor resilience. By effectively combining learning-based adaptability with the performance isolation of traditional AFTC, this approach opens new pathways in robust UAV control methodologies. The Selector-Controller structure can be generalized to other types of vehicles and fault conditions, suggesting broad potential applications.

Future developments in the AI and control systems domain may explore extending this hybrid learning approach to even more complex fault scenarios, including simultaneous multi-rotor failures. Further work could also enhance the real-time computational efficiency of the algorithms, enabling deployment on devices with stricter resource constraints. Moreover, extending the framework to accommodate predictive control strategies, especially in environments with external disturbances, might also yield beneficial outcomes.

In summary, this paper contributes to the advancement of fault-tolerant quadrotor control by demonstrating a learning-based strategy that efficiently integrates fault detection and controller selection into a single network. Its practical and theoretical implications promise significant advancements in the robustness and reliability of UAV systems under fault conditions.