Behavioral Analysis of Vision-and-Language Navigation Agents
This provides incremental insights for researchers in VLN by identifying skill-specific competencies and biases in existing agents.
The paper tackled the problem of understanding how Vision-and-Language Navigation agents ground instructions by developing a methodology to analyze their behavior on specific skills like stopping and turning, finding that biases from training affect behavior and skill-specific scores correlate with overall task performance.
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.