Event
SAE Representations Can Predict VLA Task Failure in Advance
1 reports · 1 sources
Christian A. Rogers and other authors propose a method: using sparse autoencoder (SAE) representations of OpenVLA activations to predict robot task failure. The research claims that these representations contain information predictive of outcomes and can predict task failure before the robot enters an unrecoverable state. The authors also localize individual SAE features associated with specific stages of success or failure, and identify recurring behavioral patterns through time intervals of significant activation changes; the team builds a VLM-based explanation pipeline to generate behavioral descriptions for these patterns and validates their correspondence with feature activations on held-out intervals. This work was accepted to the Science of Physical AI Safety Workshop at the 2026 Conference on Robot Learning (CoRL 2026) and is a non-archival paper.
Report timeline
- Do VLAs Know They Will Fail? Interpretable Success and Failure Patterns in Physical AI Systems
Research shows that sparse autoencoder (SAE) representations of OpenVLA activations contain outcome-predictive information and can predict task failure before the robot reaches an unrecoverable state. The authors also identify individual SAE features associated with specific stages of success or failure, and uncover recurring behavioral patterns by localizing temporal intervals where these features show significant activation changes. The team further builds a VLM-based interpretation pipeline to generate behavioral descriptions of these patterns and validates their correspondence to feature activations on held-out intervals.
arXiv cs.RO ·