FantasyVLN：面向视觉语言导航的统一多模态思维链推理
FantasyVLN: Unified Multimodal Chain-of-Thought Reasoning for Vision-and-Language Navigation
把想象出的视觉观测用预训练 VAR 编码进紧凑潜空间，使多模态思维链推理不再产生大量词元开销，从而在推理时直接做指令到动作的映射；这套「隐式推理表征」思路可迁移到任何需要推理但受延迟约束的视觉-语言-动作任务。
原文依据 imagined visual tokens are encoded into a compact latent space using a pretrained Visual AutoRegressor (VAR)
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论在视觉语言导航（VLN）中达到人类水平的表现，要求具身智能体理解文本指令、感知视觉观测，并对长动作序列进行推理。近期工作如 NavCoT 与 NavGPT-2 展示了思维链（CoT）推理在提升可解释性与长程规划方面的潜力。此外，OctoNav-R1 与 CoT-VLA 等多模态扩展进一步验证了 CoT 是通向类人导航推理的一条有前景的路径。然而，现有方法存在关键缺陷：纯文本 CoT 缺乏视觉感知，且容易过拟合于稀疏标注的推理步骤；而多模态 CoT 因生成想象出的视觉观测而带来严重的词元膨胀，使实时导航变得不切实际。在本工作中，我们提出 FantasyVLN，一个统一的隐式推理框架，在保留 CoT 推理优势的同时不引入显式的词元开销。具体而言，在 CoT 推理训练期间，想象出的视觉词元通过预训练的视觉自回归模型（VAR）被编码进紧凑的潜空间，模型在统一的多 CoT 策略下联合学习文本、视觉与多模态三种 CoT 模式。在推理时，我们的模型执行从指令到动作的直接映射，同时仍能受益于具备推理意识的表征。在 LH-VLN 上的大量实验表明，我们的方法实现了兼具推理意识与实时性的导航，在提升成功率与效率的同时，相较显式 CoT 方法将推理延迟降低了一个数量级。代码见 https://github.com/Fantasy-AMAP/fantasy-vln。
英文原文摘要 1648 字符
Achieving human-level performance in Vision-and-Language Navigation (VLN) requires an embodied agent to understand textual instructions, perceive visual observations, and reason over long action sequences. Recent works, such as NavCoT and NavGPT-2, demonstrate the potential of Chain-of-Thought (CoT) reasoning for improving interpretability and long-horizon planning. Moreover, multimodal extensions like OctoNav-R1 and CoT-VLA further validate CoT as a promising pathway toward human-like navigation reasoning. However, existing approaches face critical drawbacks: purely textual CoTs lack visual perception and easily overfit to sparse annotated reasoning steps, while multimodal CoTs incur severe token inflation by generating imagined visual observations, making real-time navigation impractical. In this work, we propose FantasyVLN, a unified implicit reasoning framework that preserves the benefits of CoT reasoning without explicit token overhead. Specifically, imagined visual tokens are encoded into a compact latent space using a pretrained Visual AutoRegressor (VAR) during CoT reasoning training, and the model jointly learns from textual, visual, and multimodal CoT modes under a unified multi-CoT strategy. At inference, our model performs direct instruction-to-action mapping while still enjoying reasoning-aware representations. Extensive experiments on LH-VLN show that our approach achieves reasoning-aware yet real-time navigation, improving success rates and efficiency while reducing inference latency by an order of magnitude compared to explicit CoT methods. Code is available at https://github.com/Fantasy-AMAP/fantasy-vln.