RC-NF: Robot-Conditioned Normalizing Flow for Real-Time Anomaly Detection in Robotic Manipulation

TL;DR

RC-NF is a real-time, unsupervised anomaly detection model for robotic manipulation that improves robustness and adaptability in dynamic environments by monitoring task states and object trajectories using normalizing flows.

Contribution

The paper introduces RC-NF, a novel robot-conditioned normalizing flow model that detects anomalies in robotic tasks using only positive samples and operates in real-time.

Findings

Achieves state-of-the-art anomaly detection accuracy across multiple categories.

Operates with a response latency under 100 ms in real-world experiments.

Enhances robustness and adaptability of VLA-based robotic systems.

Abstract

Recent advances in Vision-Language-Action (VLA) models have enabled robots to execute increasingly complex tasks. However, VLA models trained through imitation learning struggle to operate reliably in dynamic environments and often fail under Out-of-Distribution (OOD) conditions. To address this issue, we propose Robot-Conditioned Normalizing Flow (RC-NF), a real-time monitoring model for robotic anomaly detection and intervention that ensures the robot's state and the object's motion trajectory align with the task. RC-NF decouples the processing of task-aware robot and object states within the normalizing flow. It requires only positive samples for unsupervised training and calculates accurate robotic anomaly scores during inference through the probability density function. We further present LIBERO-Anomaly-10, a benchmark comprising three categories of robotic anomalies for simulation evaluation. RC-NF achieves state-of-the-art performance across all anomaly types compared to previous methods in monitoring robotic tasks. Real-world experiments demonstrate that RC-NF operates as a plug-and-play module for VLA models (e.g., pi0), providing a real-time OOD signal that enables state-level rollback or task-level replanning when necessary, with a response latency under 100 ms. These results demonstrate that RC-NF noticeably enhances the robustness and adaptability of VLA-based robotic systems in dynamic environments.