Transformer-based Localization from Embodied Dialog with Large-scale Pre-training

TL;DR

This paper introduces LED-Bert, a transformer-based model with a novel pretraining strategy for localization via embodied dialog, demonstrating that graph-based scene representations outperform traditional 2D maps and improve localization accuracy.

Contribution

The paper proposes a new LED-Bert architecture with a pretraining strategy and highlights the effectiveness of graph-based scene representations over 2D maps.

Findings

LED-Bert outperforms previous baselines

Graph-based scene representation is more effective than 2D maps

Pretraining strategy enhances localization performance

Abstract

We address the challenging task of Localization via Embodied Dialog (LED). Given a dialog from two agents, an Observer navigating through an unknown environment and a Locator who is attempting to identify the Observer's location, the goal is to predict the Observer's final location in a map. We develop a novel LED-Bert architecture and present an effective pretraining strategy. We show that a graph-based scene representation is more effective than the top-down 2D maps used in prior works. Our approach outperforms previous baselines.