Generating Contextually-Relevant Navigation Instructions for Blind and Low Vision People

TL;DR

This paper develops a dataset and explores how large pretrained language models can generate contextually relevant navigation instructions to assist blind and low-vision individuals in unfamiliar environments, validated through user studies.

Contribution

It introduces a new dataset of images and goals for navigation scenarios and demonstrates the effectiveness of language models in generating useful instructions for BLV users.

Findings

Language models produce correct, useful instructions.

User studies show perceived benefits for BLV navigation.

Insights into user preferences for different scenarios.

Abstract

Navigating unfamiliar environments presents significant challenges for blind and low-vision (BLV) individuals. In this work, we construct a dataset of images and goals across different scenarios such as searching through kitchens or navigating outdoors. We then investigate how grounded instruction generation methods can provide contextually-relevant navigational guidance to users in these instances. Through a sighted user study, we demonstrate that large pretrained language models can produce correct and useful instructions perceived as beneficial for BLV users. We also conduct a survey and interview with 4 BLV users and observe useful insights on preferences for different instructions based on the scenario.