Anomaly-Gym: A Benchmark for Anomaly Detection in Embodied Agent Environments

ICLR 2026 Conference Submission18811 Authors

Keywords: Anomaly Detection, Reinforcement Learning, Benchmark, Dataset

TL;DR: A Benchmark for Anomaly Detection for embodied Reinforcement Learning covering a range of evaluation scenarios on simulated and real-world tasks.

Abstract: Research on anomaly detection in reinforcement learning settings is sparse. Only a handful of methods have been proposed that - due to the absence of established evaluation scenarios - are evaluated on simple, small-scale, and self-proposed environments. This not only results in poor comparability but also leads to a limited understanding of the strengths and weaknesses of current approaches, rendering their applicability in real-world scenarios questionable. We address this problem by introducing Anomaly-Gym, a comprehensive evaluation suite for anomaly detection in reinforcement learning settings. In contrast to prior work, Anomaly-Gym is based on principled design criteria that disentangle evaluation from methodology. By enforcing specific constraints on the environments and anomalies considered, we propose a broad spectrum of evaluation data that covers both simulated and real-world tasks. In total, our benchmark features 10 different environments, 25 anomaly types, 4 strength levels, as well as multiple sensor modalities. We demonstrate the importance of these different aspects in a series of experiments on pre-generated datasets. For instance, we show that simple methods, while generally neglected in previous work, achieve near-perfect scores for settings with observational disturbances. In contrast, detecting perturbations of actions or environment dynamics requires more complex methods. Our findings also highlight current challenges with anomaly detection on image data and provide directions for future research.

Primary Area: datasets and benchmarks

Submission Number: 18811