Vision language action models signal task failure before breakdown
Do VLAs Know They Will Fail? Interpretable Success and Failure Patterns in Physical AI Systems
RoboticsArtificial IntelligenceMachine Learning
Summary
Vision-language-action models are robots that use images and language to decide what to do. The authors show that by looking at internal signals of these robots, it is possible to predict if a robot will fail its task before it gets stuck. They found specific signal patterns linked to success or failure and created ways to explain the robot's behavior during these times. This helps us understand why some robot attempts fail and could improve robot reliability.
What this means in practice
- •For robot control engineers: Detect early signs of robot task failure for timely intervention to prevent complete breakdowns.
- •For robotics safety teams: Identify specific robot behaviors leading to failure by interpreting signal activations for improved safety protocols.
Authors
Christian A. Rogers, Qian Lin, Daniel S. Brown
Abstract
Vision-Language-Action (VLA) models have emerged as powerful general-purpose robot policies, but their successes and failures remain difficult to diagnose and understand. In this work, we show that sparse autoencoder (SAE) representations of OpenVLA activations contain outcome-predictive information and can be used to predict task failure before the robot reaches an unrecoverable state. Moreover, to provide insights into what common behaviors may contribute to these outcomes, we identify individual SAE features whose strong activations are associated with specific stages of success or failure, and uncover recurring behavioral patterns by localizing temporal intervals where these features exhibit significant activation changes. We further develop a VLM-based interpretation pipeline to generate behavioral descriptions of these patterns and validate their correspondence to feature activations on held-out intervals.