Robotics · 2026-10-26
Adaptive World-Models for Anomaly Recognition (AWARE): A Framework for Robot Dynamics Prediction and Introspection
Luke Beddow, George Mavroghenis, Rodrigo Chacon Quesada, Kamil Dreczkowski, Cong Sun, Jiankai Wang, Oscar Kwong-Fai Pang, Antoine Cully
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
AWARE is a world-model framework for anomaly detection and introspection using only external camera monitoring. It estimates latent dynamics parameters and predicts future robot states; anomalies are detected via Mahalanobis distances on latent deviation and prediction error. On 240 real robotic crane trajectories, AWARE reports 0.93 AP (low noise) and 0.77 AP (high noise), and 88% accuracy in localizing the degrading motor.
Why it matters
The authors address passive infrastructure monitoring where onboard sensing may be unavailable or restricted. AWARE aims to maintain anomaly detection performance under noisy, camera-only state estimation by combining latent dynamics estimation with a dynamics-conditioned world model, then producing two anomaly signals for robust detection and dynamics-based fault localization.
Method
- Jointly trains a Latent Dynamics Estimator (ROA teacher–student style) to infer latent dynamics parameters from state-action history and a World Model to predict next-state conditioned on this latent.
- Uses dual anomaly scores: latent deviation (E_lat) from the reference distribution of nominal dynamics and prediction error (E_pred) from the state prediction error, both via Mahalanobis distance and calibrated thresholds.
- Evaluates architecture-agnostically with Transformer or RSSM backbones and compares against non-temporal baselines, foundation models, and related world-model/dynamics-estimation approaches.
Limitation
The authors note that they screened the dataset to ensure episodes like sensor blockage, markers, or payload leaving the field of view are absent, so the evaluation excludes these perception failure modes.
Abstract (from arXiv)
Unrecognized anomalies in critical infrastructure can rapidly cascade into catastrophic physical or operational failures. Current anomaly detection approaches often rely on direct access to the monitored system by utilizing onboard sensors, such as motor encoders, to detect anomalous states. However, legacy hardware often lacks these sensors, or strict security protocols may prohibit access altogether. We therefore present AWARE, a world model framework that is capable of anomaly detection using only external camera monitoring via system dynamics estimation and state prediction. Evaluated across 240 real-world robotic crane trajectories, including payload collisions and motor degradations, AWARE demonstrates a 26% improvement in anomaly classification over our strongest baseline under noisy, camera-only state estimation. Furthermore, the inferred dynamics parameter can be decoded to identify the source of faults, localizing the degrading motor with 88% accuracy.
Related papers
- AFTER: A World Model with Adaptive Fast--Slow Timescales for Error-Driven Retention
- World Models Dream of Success: Diagnosing and Repairing Failure Insensitivity in Robot World Models
- Acting Through Registers
- Long-WAM: Scaling the Context of World-Action Models
- OpenWAM: An Open Framework for Composable World-Action Models
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-23 · Indexed by Pub-AI: 2026-10-26 UTC
Adaptive World-Models for Anomaly Recognition (AWARE): A Framework for Robot Dynamics Prediction and Introspection
Preprint posted: 2026-10-23 · Original record ↗
- arxiv
- 2610.32742
Explicit source versions (1)
- v1 · 2026-10-23
- Preprint available
Source date: 2026-10-23 · First detected: 2026-10-27 UTC · Historical baseline
Source evidence (2)
- Official evidence ↗ · atom:published (official metadata)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Unrecognized anomalies in critical infrastructure can rapidly cascade into catastrophic physical or operational failures. Current anomaly detection approaches often rely on direct access to the monitored system by utilizing onboard sensors, such as motor encoders, to detect anomalous states. However, legacy hardware often lacks these sensors, or strict security protocols may prohibit access altogether. We therefore present AWARE, a world model framework that is capable of anomaly detection using only external camera monitoring via system dynamics estimation and state prediction. Evaluated across 240 real-world robotic crane trajectories, including payload collisions and motor degradations, AWARE demonstrates a 26% improvement in anomaly classification over our strongest baseline under noisy, camera-only state estimation. Furthermore, the inferred dynamics parameter can be decoded to identify the source of faults, localizing the degrading motor with 88% accuracy.
