Tri-Info: Generalizable, Interpretable Failure Prediction for VLA Models via Information Theory

2026-06-18

AI Summary

Vision-Language-Action (VLA) models can cause irreversible damage during physical interactions, necessitating generalizable and interpretable failure prediction mechanisms. This work formulates VLA control as a closed-loop information channel and introduces Tri-Info, an information-theoretic signal analysis method that leverages three complementary metrics—action diversity, temporal consistency, and state coupling—to enable cross-architecture, cross-environment, and even sim-to-real transferable failure detection without retraining. Evaluated across six VLA models and three benchmark environments, the proposed approach substantially outperforms existing methods, achieving an 83% accuracy in predicting failures on real-world tasks while providing interpretable diagnostic insights.

Abstract

Vision-Language-Action (VLA) models are increasingly deployed across diverse tasks, yet they remain black boxes whose physical interactions can cause irreversible harm, making generalizable and interpretable failure detection essential. We observe that successful and failed rollouts carry systematically different information-theoretic signatures. Building on this, we formalize VLA control as a closed-loop information pipeline and derive the Triple Information-theoretic (Tri-Info) signals that capture whether actions remain diverse, temporally consistent, and coupled to state transitions. Across six VLA models and three benchmark environments, Tri-Info matches the strongest baselines in-domain. Moreover, Tri-Info transfers across architectures, environments, and the sim-to-real gap without retraining, reaching 83\% accuracy on real-world tasks where prior detectors collapse to chance. This establishes Tri-Info as a simple yet powerful method that not only detects failures with strong cross-domain generalization, but also delivers interpretable diagnostics of the underlying failure modes.

Problem

• Vision-Language-Action models
• failure prediction
• generalizability
• interpretability
• information theory

Innovation

• information theory
• failure prediction
• Vision-Language-Action models
• cross-domain generalization
• interpretable AI