From Semantic Grounding to Decision Optimization: A Unified Framework for Long-Horizon UAV Vision-Language Navigation

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Zeyuan Ma, Jiaxin Chen, Di Huang
• Venue: arXiv: cs.CV
• Published: 2026-08-10
• Collected: 2026-08-11
• URL: http://arxiv.org/abs/2608.09564v1
• PDF: https://arxiv.org/pdf/2608.09564v1
• DOI: 10.1145/3767308.3836562
• Registry group: academic_apis_and_preprints
• Evidence role: canonical_academic_index
• Language: en
• Country / region: UNKNOWN

Plain-Language Brief

The authors propose a unified framework to improve UAV vision-language navigation (UAV-VLN). This framework addresses three main issues: weak grounding of instruction-relevant landmarks, insufficient exploitation of long-horizon history, and unstable decisions. The proposed solution includes an instruction-grounded semantic enhancement module, a relevance-aware dynamic temporal aggregation strategy, and a topology-aware decision method.

What the Document Says

• The study focuses on UAV vision-language navigation.
• The current methods for UAV-VLN have three main issues: weak grounding of instruction-relevant landmarks, insufficient exploitation of long-horizon history, and unstable decisions.
• A unified semantic-to-decision framework is proposed to address these issues.
• The framework includes an instruction-grounded semantic enhancement module.
• There is a relevance-aware dynamic temporal aggregation strategy in the framework.
• The framework also includes a topology-aware decision method.
• The experiments on AerialVLN and OpenFly benchmarks show that the proposed method achieves state-of-the-art performance.

Key Concepts Mentioned

• UAV vision-language navigation
• semantic enhancement module
• relevance-aware dynamic temporal aggregation
• topology-aware decision method
• AerialVLN benchmark
• OpenFly benchmark

Claims Made by the Document

• The proposed framework addresses three main issues in current UAV-VLN methods.
• The experiments demonstrate that the proposed method achieves state-of-the-art performance.

Evidence Used

• The text provides details about the proposed framework and the experiments conducted.

Methods / Approach

• Instruction-grounded semantic enhancement
• Relevance-aware dynamic temporal aggregation
• Topology-aware decision method

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

The text does not provide information about the limitations of the proposed framework.

Open Questions

• What specific techniques are used in each component of the proposed framework?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The text provides an overview of the proposed framework, but it does not go into detail about the specific techniques used in each component. Further research or examination of the original paper may provide more details.

Raw Abstract / Excerpt

UAV vision-language navigation (UAV-VLN) focuses on enabling an aerial agent to follow natural-language instructions in open 3D environments from egocentric visual observations. Current approaches suffer from three coupled issues: weak grounding of instruction-relevant landmarks in visual observations, insufficient exploitation of long-horizon history, and unstable decisions under local traps or repeated exploration. To address these issues, we propose a unified semantic-to-decision framework. First, we present an instruction-grounded semantic enhancement module that injects object-level semantics and relative spatial cues into the current observation state. Subsequently, we develop a relevance-aware dynamic temporal aggregation strategy that reweights the full history buffer while converting a few high-relevance frames into structured landmark prompts for the decoder. Finally, we devise a topology-aware decision method that combines local-optimum cognition with group-relative policy optimization under progress, goal, semantic, and path-compliance rewards. Experiments on the widely used AerialVLN and OpenFly benchmarks clearly demonstrate that our method achieves state-of-the-art performance.