Shifting Uncertainty to Critical Moments: Towards Reliable Uncertainty Quantification for VLA Model
This addresses the need for reliable introspection in VLA models to support safety-critical applications like human-in-the-loop interventions in robotics.
The paper tackles the problem of unreliable uncertainty quantification in Vision-Language-Action (VLA) models for robotics, where mean aggregation of token-level uncertainty can miss safety-critical spikes, and proposes a method that improves failure prediction accuracy on the LIBERO benchmark.
Vision-Language-Action (VLA) models enable general-purpose robotic policies by mapping visual observations and language instructions to low-level actions, but they often lack reliable introspection. A common practice is to compute a token-level uncertainty signal and take its mean over a rollout. However, mean aggregation can dilute short-lived but safety-critical uncertainty spikes in continuous control. In particular, successful rollouts may contain localized high-entropy segments due to benign noise or non-critical micro-adjustments, while failure rollouts can appear low-entropy for most timesteps and only exhibit brief spikes near the onset of failure. We propose a unified uncertainty quantification approach for predicting rollout success versus failure that (1) uses max-based sliding window pooling to preserve transient risk signals, (2) applies motion-aware stability weighting to emphasize high-frequency action oscillations associated with unstable behaviors, and (3) performs DoF-adaptive calibration via Bayesian Optimization to prioritize kinematically critical axes. Experiments on the LIBERO benchmark show that our method substantially improves failure prediction accuracy and yields more reliable signals for failure detection, which can support downstream human-in-the-loop interventions.