Behavioral Analysis of Vision-and-Language Navigation Agents

TL;DR

This paper introduces a methodology to analyze vision-and-language navigation agents by examining their grounding skills for specific actions, revealing training biases and correlating skill proficiency with overall task success.

Contribution

It presents a novel skill-specific behavioral analysis approach for VLN agents, highlighting the impact of training biases and linking skill grounding to overall performance.

Findings

Training biases affect agent behavior persistently.

Agents can ground simple referring expressions effectively.

Skill-specific scores correlate with overall VLN success.

Abstract

To be successful, Vision-and-Language Navigation (VLN) agents must be able to ground instructions to actions based on their surroundings. In this work, we develop a methodology to study agent behavior on a skill-specific basis -- examining how well existing agents ground instructions about stopping, turning, and moving towards specified objects or rooms. Our approach is based on generating skill-specific interventions and measuring changes in agent predictions. We present a detailed case study analyzing the behavior of a recent agent and then compare multiple agents in terms of skill-specific competency scores. This analysis suggests that biases from training have lasting effects on agent behavior and that existing models are able to ground simple referring expressions. Our comparisons between models show that skill-specific scores correlate with improvements in overall VLN task performance.