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

arXiv 2026 49.1 method

TLDR

DreamNav uses a coarse-to-fine diffusion-refined visual servoing with a world model for last-meter UAV precision navigation, outperforming baselines on a new benchmark.

Reasoning

The paper presents a novel framework combining trigonometric parameterization and a pre-trained world model for fine-grained spatial reasoning, supported by a large-scale benchmark (PairUAV) and zero-shot transfer results. Strengths include clear methodology and strong empirical validation; weaknesses are not explicitly discussed in the abstract but the approach appears robust.

Read-first score

Read-first score 49.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Methodology quality 18%

100

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline,benchmark,dataset,evaluation,experiment,metric

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Topical relevance 29%

64.3

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=dataset

Citation impact 18%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 705.

Keyword Scores

world model

world dynamics prediction

world simulator

generative world model

model-based reinforcement learning world model

interactive world model

video world model

Deep Analysis

Innovations

Coarse-to-fine diffusion-refined aerial visual servoing framework (DreamNav)
Trigonometric parameterization (sine/cosine) for rotation prediction to handle angular periodicity
Diffusion-refined stage using a pre-trained world model for visual imagination and action selection
PairUAV benchmark: 4.8 million image pairs across 72 scenes for last-meter UAV navigation

Methodology

DreamNav uses a two-stage approach: a coarse regression policy with trigonometric rotation parameterization, followed by a diffusion-refined stage where a pre-trained world model simulates future observations for candidate actions, selecting the trajectory that minimizes visual discrepancy with the target. Evaluation is on the new PairUAV benchmark against visual servoing and foundation model baselines, including zero-shot transfer.

Key Results

DreamNav outperforms strong visual servoing and foundation model baselines in accuracy and generalization, with zero-shot transfer to unseen scenes.

Resources

Paper