TBNAI
Speaking the Navigator's Language: Trajectory-Grounded Instruction Translation for Frozen Aerial VLN Agents
Publication: cs.AI updates on arXiv.org
Author: Xi Chen, Zhe Liu, Xiaogang Xu, Jiafei Xu, Chunyi Zhou, Yuan Su, Rui Zeng, Tianyu Du, Kelu Yao, Chao Li, Shouling Ji
Source published: 09/10/2026, 04:00:00 UTC
Briefing created: 09/10/2026, 04:36:02 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 09/10/2026, 04:03:24 UTC.
Briefings
A new method, Trajectory-Grounded Instruction Translator (TGIT), bridges the gap between user intent and frozen aerial navigation agents by converting short instructions into weak, trajectory-aligned commands. On the OpenFly navigator, this approach improved success rates from 11.33% to 37.93% using weak inputs and further boosted zero-shot transfer from 11.33% to 32.51%. The study also reports recovery performance on CityNav and AirVLN datasets.
Key points
- The TGIT method converts short, intent-driven user instructions into weak, trajectory-aligned commands to improve agent success rates.
- On the OpenFly navigator, the method increased success rate from 11.33% to 37.93% using weak inputs.
Context and limitations
The study covers only the OpenFly navigator; results on CityNav and AirVLN show recovery performance rather than direct success rates.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 09/10/2026, 04:36:02 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeid6pw2muqf5dtimzduqbkgkioexj5qzcvzxfsrat46im5rgiugivm