Last-Meter Precision Navigation for UAVs: A Diffusion-Refined Aerial Visual Servoing Approach

Authors: Yaxuan Li, Jiarui Zeng, Shaofei Huang, Zhedong Zheng

Published in arXiv preprint arXiv:2607.04352, 2026

Recommended citation: Yaxuan Li, Jiarui Zeng, Shaofei Huang, Zhedong Zheng, "Last-Meter Precision Navigation for UAVs: A Diffusion-Refined Aerial Visual Servoing Approach." arXiv preprint arXiv:2607.04352, 2026.

Code is available at: https://github.com/YaxuanLi-cn/dreamNav

Abstract: In this work, we study the last-meter precision navigation for UAVs, e.g., autonomously reaching a target within the final 10 meters using monocular vision. This task is challenging due to scale ambiguity, rotation discontinuities, and the need for fine-grained spatial reasoning. Existing methods often fail under large viewpoint changes or lack generalization to unseen environments. To this end, we propose DreamNav, a coarse-to-fine diffusion-refined aerial visual servoing framework. In the first coarse-estimation stage, a robust regression policy employs a trigonometric parameterization to predict rotation by jointly modeling sine and cosine components, effectively mitigating optimization instabilities caused by angular periodicity. Given this coarse estimate, the second diffusion-refined stage utilizes a pre-trained world model to simulate future visual observations for candidate actions, selecting the trajectory that minimizes visual discrepancy with the target through a process of visual imagination. To support rigorous evaluation, we contribute PairUAV, a large-scale benchmark comprising 4.8 million image pairs across 72 scenes, curated from the University-1652 dataset. Extensive experiments show DreamNav outperforms strong visual servoing and foundation model baselines in accuracy and generalization, with zero-shot transfer to unseen scenes.

@inproceedings{li2026last,
author = "Li, Yaxuan and Zeng, Jiarui and Huang, Shaofei and Zheng, Zhedong",
title = "Last-Meter Precision Navigation for UAVs: A Diffusion-Refined Aerial Visual Servoing Approach",
abstract = "In this work, we study the last-meter precision navigation for UAVs, e.g., autonomously reaching a target within the final 10 meters using monocular vision. This task is challenging due to scale ambiguity, rotation discontinuities, and the need for fine-grained spatial reasoning. Existing methods often fail under large viewpoint changes or lack generalization to unseen environments. To this end, we propose DreamNav, a coarse-to-fine diffusion-refined aerial visual servoing framework. In the first coarse-estimation stage, a robust regression policy employs a trigonometric parameterization to predict rotation by jointly modeling sine and cosine components, effectively mitigating optimization instabilities caused by angular periodicity. Given this coarse estimate, the second diffusion-refined stage utilizes a pre-trained world model to simulate future visual observations for candidate actions, selecting the trajectory that minimizes visual discrepancy with the target through a process of visual imagination. To support rigorous evaluation, we contribute PairUAV, a large-scale benchmark comprising 4.8 million image pairs across 72 scenes, curated from the University-1652 dataset. Extensive experiments show DreamNav outperforms strong visual servoing and foundation model baselines in accuracy and generalization, with zero-shot transfer to unseen scenes.",
booktitle = "arXiv preprint arXiv:2607.04352",
code = "https://github.com/YaxuanLi-cn/dreamNav",
year = "2026" }