Scholarlink

<p>This study investigates roll stabilization strategies for air-to-ground missiles in dynamic terminal flight, emphasizing the use of AI techniques to integrate sensor alignment with structural integrity to be benchmarked to a baseline PID control. A 6-DOF missile model served as the simulation platform. Baseline roll control was provided by PID controllers, with yaw handled via waypoint navigation and pitch via Proportional Navigation Guidance (PNG) augmented by AI models. Linear Regression (LR), Neural Networks/LSTM, Deep Deterministic Policy Gradient (DDPG) reinforcement learning, and an ensemble of these approaches were trained on PID-generated data to generate roll control inputs. Performance was assessed through error metrics (e.g., roll angle deviation) and terminal miss distance under straight flight and aggressive yaw maneuver scenarios (90° and 180° turns). The ensemble method (combining LR, LSTM, and DDPG) achieved the best results, delivering near-minimal roll error in straight flight, superior stability during sharp yaw maneuvers (90° and 180°), and the lowest miss distance of 0.14&#xa0;m in demanding scenarios. DDPG also performed robustly with consistent low miss distances and effective handling of nonlinearities and disturbances, while LR and LSTM exhibited noticeably higher roll errors and miss distances, indicating limited adaptability to rapid changes. The results demonstrate the transformative potential of AI, particularly ensemble and DDPG methods which outperformed baseline PID by enhancing precision and reducing sensitivity in challenging terminal phases. Real-world hardware validation remains essential to confirm simulation outcomes and support operational deployment.</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

AI-Augmented Roll Stabilization for Missiles Under Dynamic Terminal Flight Conditions

  • 期刊论文

  • 发表日期：

    2026年6月30日

查看全文
  • Ammar Abdurrauf,
  • Larasmoyo Nugroho,
  • Eyüp Emre Ülkü

摘要

This study investigates roll stabilization strategies for air-to-ground missiles in dynamic terminal flight, emphasizing the use of AI techniques to integrate sensor alignment with structural integrity to be benchmarked to a baseline PID control. A 6-DOF missile model served as the simulation platform. Baseline roll control was provided by PID controllers, with yaw handled via waypoint navigation and pitch via Proportional Navigation Guidance (PNG) augmented by AI models. Linear Regression (LR), Neural Networks/LSTM, Deep Deterministic Policy Gradient (DDPG) reinforcement learning, and an ensemble of these approaches were trained on PID-generated data to generate roll control inputs. Performance was assessed through error metrics (e.g., roll angle deviation) and terminal miss distance under straight flight and aggressive yaw maneuver scenarios (90° and 180° turns). The ensemble method (combining LR, LSTM, and DDPG) achieved the best results, delivering near-minimal roll error in straight flight, superior stability during sharp yaw maneuvers (90° and 180°), and the lowest miss distance of 0.14 m in demanding scenarios. DDPG also performed robustly with consistent low miss distances and effective handling of nonlinearities and disturbances, while LR and LSTM exhibited noticeably higher roll errors and miss distances, indicating limited adaptability to rapid changes. The results demonstrate the transformative potential of AI, particularly ensemble and DDPG methods which outperformed baseline PID by enhancing precision and reducing sensitivity in challenging terminal phases. Real-world hardware validation remains essential to confirm simulation outcomes and support operational deployment.