A Recurrent Vision-and-Language BERT for Navigation

TL;DR

This paper introduces a recurrent, time-aware BERT model designed for vision-and-language navigation, effectively handling partial observability and maintaining cross-modal state information to improve navigation and referring expression tasks.

Contribution

The paper proposes a novel recurrent BERT architecture tailored for VLN, enabling better history-dependent decision making and achieving state-of-the-art results.

Findings

Achieves state-of-the-art results on R2R and REVERIE datasets.

Supports pre-training and generalization to other transformer architectures.

Capable of solving navigation and referring expression tasks simultaneously.

Abstract

Accuracy of many visiolinguistic tasks has benefited significantly from the application of vision-and-language(V&L) BERT. However, its application for the task of vision-and-language navigation (VLN) remains limited. One reason for this is the difficulty adapting the BERT architecture to the partially observable Markov decision process present in VLN, requiring history-dependent attention and decision making. In this paper we propose a recurrent BERT model that is time-aware for use in VLN. Specifically, we equip the BERT model with a recurrent function that maintains cross-modal state information for the agent. Through extensive experiments on R2R and REVERIE we demonstrate that our model can replace more complex encoder-decoder models to achieve state-of-the-art results. Moreover, our approach can be generalised to other transformer-based architectures, supports pre-training, and is capable of solving navigation and referring expression tasks simultaneously.