arXiv cs.AI· Marco S. Tayar, Felipe Tommaselli, Gianluca Capezzuto, Pedro Antonio Rabelo Saraiva, Pedro H. V. de Freitas, Lucas Kido, Guilherme Sonego, Ricardo V. Godoy, Marcelo Becker·· 14 小时前AI 评分38
VLN on the Fly：面向空中机器人的机载视觉语言导航栈
VLN on the Fly: An Onboard Vision-Language Navigation Stack for Aerial Robots
AI 导读
研究团队提出 VLN on the Fly，一套将指令 grounding、规划与控制保留为独立可检查阶段的机载视觉语言导航栈，由量化 VLM 将指令定位到图像粗网格、深度信息抬升为 3D 目标、快速 B 样条规划器生成可行轨迹，再由预训练强化学习策略输出电机指令。
正文 · 原文
Abstract:Running vision-language navigation fully onboard an aerial robot is hard, since grounding, planning, and control must share limited compute and a single-stage error is difficult to isolate in flight. End-to-end aerial policies fuse these stages into one network, giving up the observability and safety checks a modular stack keeps available. We propose VLN on the Fly, an onboard stack that keeps grounding, planning, and control as separate, inspectable stages. A quantized VLM grounds an instruction to a coarse image cell, depth lifts it to a 3D goal, a fast B-spline planner returns a feasible trajectory, and a pretrained reinforcement learning policy tracks it to motor commands across quadrotors. Across 15 onboard flights over three everyday referents in a controlled indoor volume, the stack reaches the target in 13 of 15 trials with 5.72 cm mean goal error and 39.3% average GPU utilization. In 6 additional cluttered-environment trials, the stack tracks collision-free trajectories under onboard perception gating.
| Comments: | 7 pages, 5 figures. Project page: this https URL |
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Systems and Control (eess.SY) |
| Cite as: | arXiv:2609.20191 [cs.RO] |
| (or arXiv:2609.20191v2 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2609.20191 arXiv-issued DOI via DataCite |
Submission history
From: Gianluca Capezzuto [view email]
[v1]
Fri, 7 Aug 2026 20:34:33 UTC (7,770 KB)
[v2]
Sun, 25 Oct 2026 00:46:44 UTC (7,772 KB)
来源：arXiv cs.AI · arxiv.org