Tri-Info: Generalizable, Interpretable Failure Prediction for VLA Models via Information Theory
Provides a generalizable and interpretable failure detection method for VLA models, addressing a critical safety need in robotics.
The authors propose Tri-Info, an information-theoretic failure detection method for VLA models that achieves 83% accuracy on real-world tasks without retraining, outperforming prior methods that collapse to chance.
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.