FantasyVLN: Unified Multimodal Chain-of-Thought Reasoning for Vision-Language Navigation

TL;DR

FantasyVLN introduces a unified multimodal reasoning framework for vision-language navigation that maintains interpretability and reasoning capabilities without token inflation, enabling real-time, human-like navigation performance.

Contribution

The paper proposes FantasyVLN, a novel implicit reasoning approach that encodes imagined visual tokens into a compact space, reducing token inflation and improving real-time navigation efficiency.

Findings

Achieves higher success rates in LH-VLN benchmark.

Reduces inference latency by an order of magnitude.

Maintains reasoning capabilities without explicit token overhead.

Abstract

Achieving human-level performance in Vision-and-Language Navigation (VLN) requires an embodied agent to jointly understand multimodal instructions and visual-spatial context while reasoning 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 spatial grounding 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.