---
title: "FSD-VLN: Fast-Slow Dual-System Modeling for Aerial Long-Horizon Vision-Language Navigation"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.08359v1"
permalink: "https://feed.astralor.com/items/itm_b95fb85acdb34d1c87565e2c"
published: "2026-07-09T11:17:04.000Z"
score: 76
channel: "tech"
tags: ["FSD-VLN","Fast-Slow System","Diffusion Transformer","UAV Autonomous Navigation","Vision-Language Navigation"]
---

# FSD-VLN: Fast-Slow Dual-System Modeling for Aerial Long-Horizon Vision-Language Navigation

## AI commentary

> This paper offers an elegant paradigm shift for solving action latency in long-horizon drone navigation by borrowing the fast-slow system from cognitive science, rather than just stacking larger models. For readers in robotics, vision-language navigation, or embedded AI, the time-aware optimizer and DiT application deserve deep study. It also reminds us that decoupling can sometimes be the key to real-time intelligence.

## At a glance

This paper proposes FSD-VLN, a fast-slow dual-system architecture to address the structural misalignment between global multimodal understanding and low-latency action generation in long-horizon aerial vision-language navigation. The framework comprises a slow stream that extracts stable semantic priors from pre-trained VLMs and a fast stream using Diffusion Transformer to model cross-temporal action distributions, along with a time-aware adaptive optimizer for stable long-sequence training. Large-scale low-altitude simulations show up to 2× higher navigation success rates and over 50% reductions in per-action inference delay and total task runtime, validating the benefit of decoupled semantic-control modeling.

## Key points

- Proposes a fast-slow dual-system architecture that decouples semantic reasoning from low-latency flight control.
- The fast stream uses a Diffusion Transformer (DiT) to generate consistent cross-temporal action sequences.
- Introduces a time-aware adaptive optimizer to reduce gradient oscillation during long-sequence training.
- Achieves up to 2× higher navigation success rates on unseen scenes and over 50% reduction in delay compared to SOTA methods.

## Full text

Vision-Language Navigation (VLN) enables UAV autonomous navigation in unknown environments by mapping language instructions to real-time visual inputs. Compared with GPS-dependent or pre-programmed navigation, VLN supports intuitive human-machine interaction and stronger environmental adaptability, requiring tight integration of high-level semantic reasoning and low-latency flight control.Existing methods suffer from structural misalignment between global multimodal understanding and sequential action generation, causing jittery trajectories and severe decision latency for long-horizon aerial navigation. To solve this issue, we propose FSD-VLN, a fast-slow dual-system architecture disentangling semantic reasoning and low-latency flight command generation.The framework has two asynchronous branches: a slow stream extracting stable semantic priors from pre-trained vision-language models, and a Diffusion Transformer (DiT) fast stream modeling cross-temporal action distributions to produce consistent flight outputs. We further introduce a time-aware adaptive optimizer to stabilize long-sequence training and reduce gradient oscillation.Large-scale low-altitude simulation experiments show FSD-VLN achieves up to 2X higher navigation success rates on unseen scenes than SOTA methods, while cutting single-action inference delay and total task runtime by over 50%. Our work validates the benefit of decoupled semantic-control modeling and provides a practical paradigm for long-horizon aerial VLN.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_b95fb85acdb34d1c87565e2c