# X-Fly：学习在 Full-Chunk 空中视觉-语言导航中抵达并停止

> 原文链接：https://robotics.hellobike.com/papers/2026-10-15-2610-21450
> 频道：papers | 更新日期：2026-10-15 | 标签：arxiv, cs.RO
> 核心摘要：X-Fly针对完整动作块空中VLN的恢复与停止问题，用分阶段训练在OpenFly测试集验证性能提升。

# X-Fly：学习在 Full-Chunk 空中视觉-语言导航中抵达并停止

::: v-pre
> 原文链接：[X-Fly: Learning to Reach and Stop in Full-Chunk Aerial Vision-Language Navigation](https://arxiv.org/abs/2610.21450)（arXiv 2610.21450 · 发布日期：2026-10-15）

**X-Fly: Learning to Reach and Stop in Full-Chunk Aerial Vision-Language Navigation**

X-Fly针对完整动作块空中VLN的恢复与停止问题，用分阶段训练在OpenFly测试集验证性能提升。

## 核心要点

- 完整动作块可摊销大模型查询成本，但使查询状态偏离专家示范并暴露停止失败。
- Recovery DAgger为学习者状态标注重连专家轨迹的平滑块及首个区域内STOP。
- Flow-GRPO用同任务闭环结果联合优化块式流控制器和Bernoulli STOP策略。
- 每次决策八个控制时，X-Fly在test-seen/test-unseen达74.18%/51.73% SR。

**验证范围与局限**：验证限于OpenFly的test-seen和test-unseen设置，摘要未说明真机实验。

## 英文摘要 (Abstract)

Aerial vision-language navigation (VLN) agents built on large vision-language models are expensive to query. Executing a complete predicted action chunk before replanning amortizes this cost, reducing model inference from 195 to 42 ms per executed waypoint for our policy. It also moves the states at which the policy is queried away from expert demonstrations, which supervise neither recovery from these states nor termination at the learner's own arrival. On OpenFly test-seen, no-stop failures account for 47.78% of full-chunk behavior cloning (BC) outcomes. We present X-Fly, a staged training framework for full-chunk aerial VLN. Recovery DAgger labels learner-visited decision states with smooth chunks that reconnect to future expert segments and with STOP at the first in-region state. Flow-GRPO then jointly refines the chunked flow controller and a Bernoulli STOP policy from same-task closed-loop outcomes, supplemented by outcome-verified stopping supervision. With eight controls per decision, X-Fly reaches 74.18% SR on test-seen and 51.73% on test-unseen, compared with 16.45% and 18.43% for full-chunk BC and 32.04% and 29.46% for DreamFly.

作者：Ruibin Ju, Ruifeng Cai, Min Xu, Lin Wang
:::

## 原始参考来源 (Sources)

- [arXiv 2610.21450：X-Fly: Learning to Reach and Stop in Full-Chunk Aerial Vision-Language Navigation](https://arxiv.org/abs/2610.21450)