Scholarlink

<p>In this work, to address safety concerns in densely populated urban areas, a human-in-loop deep deterministic policy gradient (HIL-DDPG) model is developed for collision-aware path planning in GPS-directed autonomous drones. The methodology combined reinforcement learning with human intervention, utilizing training data from simulated and real-world flight scenarios across diverse environments. Analysis revealed that the HIL-DDPG architecture employing deep reinforcement learning to identify optimal routes while continuously monitoring collision probabilities, performs well in dense urban scenarios. When safety thresholds are compromised, the system transfers control to a human operator, effectively integrating machine expertise with human judgment. Experimental results demonstrated approximately 80 % fewer collision incidents while maintaining computational efficiency. The system successfully navigated challenging urban conditions and adverse weather conditions with minimal human assistance, enhancing autonomous drone reliability for safety-critical applications including search and rescue operations, delivery services, urban surveillance and military missions.</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

RL-based path optimization for GPS-navigated drones using human-in-loop deep deterministic policy gradient technique

  • 期刊论文

  • 发表日期：

    2026年8月5日

查看全文
  • Bermagambet Duisek,
  • Al-Adi Zhunus,
  • Muhammad Ilyas

摘要

In this work, to address safety concerns in densely populated urban areas, a human-in-loop deep deterministic policy gradient (HIL-DDPG) model is developed for collision-aware path planning in GPS-directed autonomous drones. The methodology combined reinforcement learning with human intervention, utilizing training data from simulated and real-world flight scenarios across diverse environments. Analysis revealed that the HIL-DDPG architecture employing deep reinforcement learning to identify optimal routes while continuously monitoring collision probabilities, performs well in dense urban scenarios. When safety thresholds are compromised, the system transfers control to a human operator, effectively integrating machine expertise with human judgment. Experimental results demonstrated approximately 80 % fewer collision incidents while maintaining computational efficiency. The system successfully navigated challenging urban conditions and adverse weather conditions with minimal human assistance, enhancing autonomous drone reliability for safety-critical applications including search and rescue operations, delivery services, urban surveillance and military missions.