Scholarlink

To enhance guidance accuracy in near-space target interception, this paper presents an onboard deployment scheme of an attention-enhanced LSTM encoder network for real-time trajectory prediction. The system integrates Kalman filtering for motion reconstruction and maneuver recognition, enabling multi-modal prediction under complex target dynamics. A lightweight version of the model is deployed on a missile-borne ARM + Atlas200 + FPGA heterogeneous inference module, maintaining accuracy degradation within 10% after pruning. Hardware-in-the-loop (HIL) validation demonstrates that the proposed system achieves a trajectory forecast error within 10 km over a 60-second prediction horizon. The results validate the effectiveness of deploying deep learning-based trajectory prediction intelligence directly onboard missile platforms, providing a practical solution to address latency and uncertainty in mid-course guidance scenarios.

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

Onboard Deployment of Trajectory Prediction Intelligence HIL Validation for Missile Guidance Systems

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Tao Yan,
  • Ya Yang,
  • Wei Yin

摘要

To enhance guidance accuracy in near-space target interception, this paper presents an onboard deployment scheme of an attention-enhanced LSTM encoder network for real-time trajectory prediction. The system integrates Kalman filtering for motion reconstruction and maneuver recognition, enabling multi-modal prediction under complex target dynamics. A lightweight version of the model is deployed on a missile-borne ARM + Atlas200 + FPGA heterogeneous inference module, maintaining accuracy degradation within 10% after pruning. Hardware-in-the-loop (HIL) validation demonstrates that the proposed system achieves a trajectory forecast error within 10 km over a 60-second prediction horizon. The results validate the effectiveness of deploying deep learning-based trajectory prediction intelligence directly onboard missile platforms, providing a practical solution to address latency and uncertainty in mid-course guidance scenarios.