RAD: A Dataset and Benchmark for Real-Life Anomaly Detection with Robotic Observations

2026-07-27 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

The Realistic Anomaly Detection (RAD) dataset is introduced to address the critical lack of realistic data for industrial anomaly detection. Collected using a Franka Emika Panda robotic arm equipped with an Intel RealSense D415 camera, RAD comprises 4765 images across 13 object categories and 4 distinct defect types: scratched, missing, stained, and squeezed. The dataset features observations from over 50 viewpoints, specifically designed to mirror real-world industrial conditions, including environmental noise, variable object poses, and unstable camera positions, which often hinder existing anomaly detection algorithms. Additionally, the paper proposes a data augmentation method to improve pose estimation accuracy and facilitate 3D point cloud reconstruction, systematically evaluating state-of-the-art RGB-based and point cloud models to identify current limitations and future research directions.

Key takeaway

For Machine Learning Engineers developing industrial anomaly detection systems, this work highlights that current models often underperform in real-world robotic environments due to noise and pose inaccuracies. You should prioritize evaluating your algorithms on realistic multi-view datasets like RAD and consider integrating robust pose estimation and 3D reconstruction pipelines to enhance model reliability and effectiveness in production settings.

Key insights

RAD dataset bridges the gap between idealized and real-world industrial anomaly detection by incorporating robotic observation noise.

Principles

• Real-world noise significantly impacts anomaly detection performance.
• Multi-view data is essential for comprehensive 3D anomaly detection.
• Accurate camera pose estimation is critical for 3D reconstruction.

Method

A pipeline for self-supervised pose estimation and 3D point cloud reconstruction is proposed, involving distortion, pose, and shape estimation, enhanced by image preprocessing and compactness comparison for accuracy.

In practice

• Utilize robotic arms for consistent, efficient multi-view data collection.
• Implement image preprocessing (brightness, contrast, gamma) for robust pose estimation.
• Assess model performance across diverse defect types and noisy conditions.

Topics

• Anomaly Detection
• Robotics
• Multi-view Imaging
• Dataset Benchmark
• Pose Estimation
• 3D Reconstruction

Code references

• kaichen-z/PAD
• kaichen-z/RAD

Best for: Computer Vision Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, Robotics Engineer

Related on AIssential

• Breaking the Rigid Prior: Towards Articulated 3D Anomaly Detection — Articulated 3D anomaly detection requires disentangling pose-induced geometry from true structural defects.
• SwinAD: Multi-stage feature reconstruction for unsupervised industrial anomaly detection — SwinAD uses a Swin Transformer V2 encoder and diverse multi-scale reconstruction to improve pixel-level anomaly localization in multi-class industrial settings.
• Anomaly Factory 3D: A Modular Framework for Diverse Pseudo-Anomaly Synthesis in Unsupervised 3D Anomaly Detection — AF3AD synthesizes diverse 3D pseudo-anomalies using a parametric deformation model to enhance unsupervised anomaly detection training.
• Bounding-Box Trajectories Matter for Video Anomaly Detection — Bounding-box trajectories, a byproduct of detection and tracking, are a powerful, multi-class, and efficient primary signal for video anomaly detection.
• Visual Prompting Meets Feature Reconstruction-Based Anomaly Detection with Dual-Teacher Supervision — Robust anomaly detection in varied real-world conditions is achieved by combining visual prompting, adaptive teacher models, and synthetic data.
• OPD-IAD: From Language Judgment to Industrial Anomaly Detection via On-Policy Self-Distillation — OPD-IAD precisely localizes industrial anomalies by distilling language judgments onto visual features via on-policy self-distillation and visual anchoring.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.