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

2026-07-09

AI Summary

This work addresses the challenges of trajectory jitter and decision latency in existing vision-language navigation methods for long-horizon aerial tasks, which stem from the tight coupling between semantic understanding and action generation. To overcome this, the authors propose a fast-slow dual-system architecture that decouples high-level semantic reasoning from low-latency flight control through an asynchronous dual-stream mechanism, enabling stable and efficient navigation. The approach integrates a pretrained vision-language model to extract semantic priors and employs a Diffusion Transformer to model cross-temporal action distributions. Additionally, a time-aware adaptive optimizer is introduced to enhance training stability and navigational consistency over long sequences. Evaluated on large-scale low-altitude simulations, the method achieves up to a 2× improvement in navigation success rate over the best prior approach, while reducing both per-step inference latency and total task duration by more than 50%.

Abstract

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.

Problem

• Vision-Language Navigation
• Aerial Navigation
• Long-Horizon Navigation
• Decision Latency
• Trajectory Jitter

Innovation

• Fast-Slow Dual-System
• Vision-Language Navigation
• Diffusion Transformer
• Aerial Navigation
• Time-Aware Optimization