Event
SentryVLA: A Runtime Reliability Wrapper for VLA Without Retraining
1 reports · 1 sources
Zongyao Lyu, Lingling Chen, and William J. Beksi propose SentryVLA, a lightweight runtime wrapper for frozen vision-language-action (VLA) policies that requires no model retraining or architectural changes. At the action level, it uses conformal quantile regression to estimate an upper bound on the deviation between sampled actions and expert behavior, and on that basis ranks multiple stochastic candidate actions; at the execution level, it uses a Mahalanobis-distance anomaly score to monitor anomalous states in closed-loop control. The authors report that simulation and real-world experiments show the method improves action selection and provides useful runtime warning signals for anomalous execution states.
Report timeline
- SentryVLA: Uncertainty-Guided Action Selection and State Monitoring for Vision-Language-Action Robotic Control
SentryVLA is a lightweight runtime wrapper for frozen VLA policies that requires no retraining or changes to the model architecture. At the action level, it uses conformal quantile regression to estimate an upper bound on the deviation between sampled actions and expert behavior, and ranks multiple stochastic candidate actions accordingly; at the execution level, it uses a Mahalanobis-distance anomaly score to monitor abnormal states during closed-loop control. Simulation and real-world experiments show that the method improves action selection and provides useful runtime warning signals.
arXiv cs.RO ·