ME-AD: Progressive Robotics Anomaly Detection Dataset 机器收录·待核验

mxds0xf6z36k3v11

Introduction

The ME-AD (Mitsubishi Electric Anomaly Detection) dataset contains operating data of cyclic pick-and-place executions collected from a 6-DoF Mitsubishi Electric RV-7FM-D1-S15 manipulator under progressive actuation degradation. A defect was intentionally introduced in the actuator of joint 3 (removal of gearbox lubricant), causing gradual mechanical wear over pick-and-place cycles. Unlike existing robotics AD datasets where anomalies are artificially induced, ME-AD documents an actual progressive hardware fault, making it uniquely suited for benchmarking semi-supervised anomaly detection, online/early fault detection, and remaining useful life estimation.

At a Glance

The size of the unzipped dataset is ~10GB. The dataset contains approximately 17,000 pick-and-place cycles organized into 7 operation groups called tasks across 3 motion families. Each cycle includes joint positions, velocities, accelerations, and torques sampled at ~280 Hz.

The dataset provides raw CSV files, cleaned Pandas DataFrames (.pkl), and pre-configured AD task splits. Each data sample contains 6 joint positions (rad), 6 joint velocities (rad/s), 6 joint accelerations (rad/s²), and 6 joint torques (Nm). Processed files include both raw and low-pass filtered signals.

ME-AD/ contains: CSV/ (raw per-cycle CSV files organized by operation code), Pandas/ (cleaned .pkl DataFrames per cycle), Tasks/ (7 pre-configured AD benchmark tasks with training/healthy/faulty splits), and Python scripts for data processing and task generation.

The README.md file contains further details.

Citation

If you use the ME-AD dataset in your research, please cite our contribution:

@article{giacomuzzo2026anomaly,

title={Anomaly Detection in Industrial Robotics: A Novel Dataset and Benchmark Method},

author={Giacomuzzo, Giulio and Dalla Libera, Alberto and Arimizu, Daichi and Hachiya, Yuto and Tsuruta, Shinya and Wakimoto, Koji and Kuriyama, Toshiyuki and Romeres, Diego},

journal={},

year={2

本卡描述如含来自官方页的原文片段,其版权归原作者,不在本站 CC-BY 4.0 许可范围(见关于与许可)。

落地页

https://zenodo.org/records/20817531

许可证

CC-BY-SA-4.0 (置信:verified_official)

发布年份

2026

发布方

Zenodo

分发点