Diagnosing Vision-and-Language Navigation: What Really Matters
This work addresses the unclear inner mechanisms of navigation agents for researchers in multimodal AI, though it is incremental as it diagnoses existing models rather than proposing new ones.
The paper tackled the problem of understanding how agents perceive multimodal input in vision-and-language navigation, finding that indoor agents use both object and direction tokens while outdoor agents rely heavily on direction tokens, and Transformer-based agents show better cross-modal understanding and numerical reasoning.
Vision-and-language navigation (VLN) is a multimodal task where an agent follows natural language instructions and navigates in visual environments. Multiple setups have been proposed, and researchers apply new model architectures or training techniques to boost navigation performance. However, there still exist non-negligible gaps between machines' performance and human benchmarks. Moreover, the agents' inner mechanisms for navigation decisions remain unclear. To the best of our knowledge, how the agents perceive the multimodal input is under-studied and needs investigation. In this work, we conduct a series of diagnostic experiments to unveil agents' focus during navigation. Results show that indoor navigation agents refer to both object and direction tokens when making decisions. In contrast, outdoor navigation agents heavily rely on direction tokens and poorly understand the object tokens. Transformer-based agents acquire a better cross-modal understanding of objects and display strong numerical reasoning ability than non-Transformer-based agents. When it comes to vision-and-language alignments, many models claim that they can align object tokens with specific visual targets. We find unbalanced attention on the vision and text input and doubt the reliability of such cross-modal alignments.