Why Only Text: Empowering Vision-and-Language Navigation with Multi-modal Prompts

TL;DR

This paper introduces VLN-MP, a new task that enhances vision-and-language navigation by integrating multi-modal prompts, including images, to reduce ambiguity and improve agent performance across various benchmarks.

Contribution

The paper proposes VLN-MP, a novel multi-modal prompt framework for VLN, along with a new benchmark, enabling better integration of visual signals and improved navigation accuracy.

Findings

Visual prompts significantly improve navigation performance.

VLN-MP outperforms text-only models across benchmarks.

The framework maintains efficiency with text prompts.

Abstract

Current Vision-and-Language Navigation (VLN) tasks mainly employ textual instructions to guide agents. However, being inherently abstract, the same textual instruction can be associated with different visual signals, causing severe ambiguity and limiting the transfer of prior knowledge in the vision domain from the user to the agent. To fill this gap, we propose Vision-and-Language Navigation with Multi-modal Prompts (VLN-MP), a novel task augmenting traditional VLN by integrating both natural language and images in instructions. VLN-MP not only maintains backward compatibility by effectively handling text-only prompts but also consistently shows advantages with different quantities and relevance of visual prompts. Possible forms of visual prompts include both exact and similar object images, providing adaptability and versatility in diverse navigation scenarios. To evaluate VLN-MP under a unified framework, we implement a new benchmark that offers: (1) a training-free pipeline to transform textual instructions into multi-modal forms with landmark images; (2) diverse datasets with multi-modal instructions for different downstream tasks; (3) a novel module designed to process various image prompts for seamless integration with state-of-the-art VLN models. Extensive experiments on four VLN benchmarks (R2R, RxR, REVERIE, CVDN) show that incorporating visual prompts significantly boosts navigation performance. While maintaining efficiency with text-only prompts, VLN-MP enables agents to navigate in the pre-explore setting and outperform text-based models, showing its broader applicability.