Scholarlink

<p>The increasing deployment of flexible satellite structures, such as solar panels and antennas, introduces significant control challenges due to elastic vibrations and underactuation caused by partial actuator faults. This paper proposes a hybrid attitude control framework that integrates robust nonlinear controllers, super-twisting algorithm (STA), and adaptive higher-order sliding mode control (AHSMC), with deep deterministic policy gradient (DDPG), a state-of-the-art reinforcement learning algorithm. The DDPG agent autonomously tunes the control gains in real time, optimizing the system's tracking accuracy and energy efficiency in the presence of structural uncertainties and dynamic disturbances. The proposed framework is evaluated through both simulation and real-time implementation on rigid and flexible satellite models. Results demonstrate that the RL-augmented controllers significantly improve attitude tracking performance, reduce chattering effects, and enhance fault tolerance, particularly in underactuated configurations. Among the tested strategies, the STA–DDPG combination yields the best trade-off between robustness, settling time, and mean squared error. These findings confirm the efficacy and adaptability of the proposed method for next-generation autonomous satellite systems operating in complex orbital environments.</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

Deep Reinforcement Learning-Enhanced Attitude Control of Underactuated Flexible Satellites

  • 期刊论文

  • 发表日期：

    2026年7月8日

查看全文
  • Jamal Zarei,
  • Maryam Malekzadeh

摘要

The increasing deployment of flexible satellite structures, such as solar panels and antennas, introduces significant control challenges due to elastic vibrations and underactuation caused by partial actuator faults. This paper proposes a hybrid attitude control framework that integrates robust nonlinear controllers, super-twisting algorithm (STA), and adaptive higher-order sliding mode control (AHSMC), with deep deterministic policy gradient (DDPG), a state-of-the-art reinforcement learning algorithm. The DDPG agent autonomously tunes the control gains in real time, optimizing the system's tracking accuracy and energy efficiency in the presence of structural uncertainties and dynamic disturbances. The proposed framework is evaluated through both simulation and real-time implementation on rigid and flexible satellite models. Results demonstrate that the RL-augmented controllers significantly improve attitude tracking performance, reduce chattering effects, and enhance fault tolerance, particularly in underactuated configurations. Among the tested strategies, the STA–DDPG combination yields the best trade-off between robustness, settling time, and mean squared error. These findings confirm the efficacy and adaptability of the proposed method for next-generation autonomous satellite systems operating in complex orbital environments.