基于目标中心奖励的开放世界无人机主动跟踪
摘要
无人机视觉主动跟踪旨在通过基于视觉观察来控制运动系统的运动，以在动态环境中实现有效的跟踪，这为实际应用中的有效跟踪提供了更实用的解决方案。然而，由于缺乏统一的基准和开放世界环境中频繁干扰的复杂性，使用强化学习实现准确的无人机视觉主动跟踪仍然具有挑战性。为了解决这些问题，我们提出了一套系统化的解决方案。首先，我们提出了DAT，即首个开放世界无人机主动空中-地面跟踪基准。它包含24个城市规模场景， featuring targets with human-like behaviors and high-fidelity dynamics simulation. DAT also provides a digital twin tool for unlimited scene generation. Additionally, we propose a novel reinforcement learning method called GC-VAT, which aims to improve the performance of drone tracking targets in complex scenarios. Specifically, we design a Goal-Centered Reward to provide precise feedback across viewpoints to the agent, enabling it to expand perception and movement range through unrestricted perspectives. Inspired by curriculum learning, we introduce a Curriculum-Based Training strategy that progressively enhances the tracking performance in complex environments. Besides, experiments on simulator and real-world images demonstrate the superior performance of GC-VAT, achieving a Tracking Success Rate of approximately 72% on the simulator. The benchmark and code are available at https://github.com/SHWplus/DAT_Benchmark.
引用
@article{arxiv.2412.00744,
title = {Open-World Drone Active Tracking with Goal-Centered Rewards},
author = {Haowei Sun and Jinwu Hu and Zhirui Zhang and Haoyuan Tian and Xinze Xie and Yufeng Wang and Xiaohua Xie and Yun Lin and Zhuliang Yu and Mingkui Tan},
journal= {arXiv preprint arXiv:2412.00744},
year = {2025}
}
备注
NeurIPS 2025