A Monocular UAV Interception Method Based on Reinforcement Learning and End-to-End Control
赵帅平, Ye Yang, Liaoni Wu
Xiamen University
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
This paper studies an autonomous interception method for a monocular quadrotor unmanned aerial vehicle (UAV) against dynamic targets. Considering the limited information available from monocular vision and the frequency mismatch between target perception and UAV control, a hierarchical control framework is proposed that includes a reinforcement learning (RL) teacher policy and an end-to-end vision student policy. First, a monocular visual interception task for the UAV is built in a 3D simulation environment. Then, low-dimensional observations are defined using the UAV's velocity and the target's line-of-sight (LOS) angle deviation, and the Deep Deterministic Policy Gradient (DDPG) algorithm is adopted to train the UAV's autonomous guidance policy. Subsequently, this policy is utilized to generate flight sample data, and an end-to-end network based on ResNet18 is designed to achieve a direct mapping from monocular images to UAV control commands. Finally, the model is deployed to the ROS2/PX4 system to accomplish online closed-loop control. The results show that the proposed method successfully achieves the transfer from RL guidance to monocular vision end-to-end control, verifying its feasibility in UAV dynamic target interception tasks.
逐年被引趋势
暂无年度引用数据
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
工程Guidance and Control Systems
Aerospace Engineering and Control Systems · Reinforcement Learning in Robotics