- Updated: July 16, 2026
- 2 min read
FSD‑VLN: Fast‑Slow Dual‑System Modeling for Aerial Long‑Horizon Vision‑Language Navigation
FSD‑VLN: Fast‑Slow Dual‑System Modeling for Aerial Long‑Horizon Vision‑Language Navigation
Authors: Xueke Zhu, Qingyan Meng, Liutao Yu, Wei Zhang, Zhengyu Ma, Huihui Zhou, Yonghong Tian
Abstract: Vision‑Language Navigation (VLN) enables UAVs to autonomously navigate unknown environments by translating natural‑language instructions into real‑time visual actions. Existing approaches suffer from a structural mismatch between global multimodal understanding and sequential action generation, leading to jittery trajectories and high decision latency for long‑horizon aerial tasks.
In this work we introduce FSD‑VLN, a fast‑slow dual‑system architecture that decouples semantic reasoning from low‑latency flight command generation. The slow stream extracts stable semantic priors using pretrained vision‑language models, while a Diffusion Transformer (DiT) fast stream models cross‑temporal action distributions to produce consistent flight outputs. A time‑aware adaptive optimizer further stabilises long‑sequence training and reduces gradient oscillation.
Extensive low‑altitude simulation experiments demonstrate up to 2× higher navigation success rates on unseen scenes compared to state‑of‑the‑art methods, while cutting single‑action inference delay and total task runtime by more than 50 %. These results validate the benefit of decoupled semantic‑control modelling for long‑horizon aerial VLN.
Key Contributions
- Dual‑system architecture separating semantic priors (slow) from rapid flight command generation (fast).
- Diffusion Transformer (DiT) for modelling cross‑temporal action distributions.
- Time‑aware adaptive optimizer for stable long‑sequence training.
- Significant performance gains on large‑scale aerial VLN benchmarks.
Read the full paper on arXiv and explore related resources on our site: ubos.tech/blog.
Andrii Bidochko
CTO UBOS
Andrii Bidochko is an AI entrepreneur and researcher focused on AI agents, reinforcement learning, and autonomous systems. He writes about the technologies shaping the future of machine intelligence, from frontier models and agent architectures to real-world AI applications.