Scholarlink

<p>Quadrotor UAV attitude control faces challenges posed by strong nonlinearity, model uncertainty, and external disturbances. Traditional neural network adaptive control primarily relies on the network’s inherent memory to handle sudden command changes, enhancing adaptation speed by increasing the learning rate. To address these issues, this paper proposes an innovative adaptive attitude control strategy that combines Memory Augmented Neural Network with Non-Singular Terminal Sliding Mode Control. Within this control architecture, NTSMC provides robust baseline control, ensuring finite-time convergence of tracking errors while avoiding singularities. Concurrently, MANN enhances learning efficiency through an external working memory mechanism. MANN estimates are integrated into NTSMC as feedforward compensation, enabling rapid adaptation and compensation for unknown nonlinear dynamics and disturbances. Furthermore, the proposed control method introduces dynamically adaptive control gains to prevent overestimation while ensuring the system state reaches the terminal sliding mode manifold, thereby further reducing chatter. Lyapunov stability analysis demonstrates the unified boundedness and stability of the closed-loop system. The approach is ultimately validated in a simulation environment incorporating time-varying wind disturbances and multiple reference trajectories. Results indicate significant advantages in tracking accuracy, convergence speed, and disturbance suppression performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Adaptive attitude control of quadrotor UAV based on memory augmented neural network

  • 期刊论文

  • 发表日期：

    2026年3月24日

查看全文
  • Zhanshan Zhao,
  • Wenhui Guo,
  • Guowei Xu,
  • Jiao Yin,
  • Jinli Cao,
  • Hua Wang

摘要

Quadrotor UAV attitude control faces challenges posed by strong nonlinearity, model uncertainty, and external disturbances. Traditional neural network adaptive control primarily relies on the network’s inherent memory to handle sudden command changes, enhancing adaptation speed by increasing the learning rate. To address these issues, this paper proposes an innovative adaptive attitude control strategy that combines Memory Augmented Neural Network with Non-Singular Terminal Sliding Mode Control. Within this control architecture, NTSMC provides robust baseline control, ensuring finite-time convergence of tracking errors while avoiding singularities. Concurrently, MANN enhances learning efficiency through an external working memory mechanism. MANN estimates are integrated into NTSMC as feedforward compensation, enabling rapid adaptation and compensation for unknown nonlinear dynamics and disturbances. Furthermore, the proposed control method introduces dynamically adaptive control gains to prevent overestimation while ensuring the system state reaches the terminal sliding mode manifold, thereby further reducing chatter. Lyapunov stability analysis demonstrates the unified boundedness and stability of the closed-loop system. The approach is ultimately validated in a simulation environment incorporating time-varying wind disturbances and multiple reference trajectories. Results indicate significant advantages in tracking accuracy, convergence speed, and disturbance suppression performance.