← Back ICRA 2026

Contrastive Learning on 3D Point Clouds for Robotic Geometric Defect Detection

Alexander Tarvo, Yusen Wan, Xu Chen

PDF

AI summary

Key figure (auto-extracted from paper)

COSARAD achieves state-of-the-art geometric defect detection on 3D point clouds by using contrastive learning and location-specific memory banks to eliminate spatial ambiguity.

3D point clouds anomaly detection contrastive learning robotic inspection geometric defect detection spatial awareness

Problem

Existing 3D defect detection methods suffer from spatial indifference, where monolithic feature banks ignore coordinate context, causing high false-positive rates and inconsistent detection of geometric anomalies.

Approach

The method trains a contrastive feature extractor on 3D point cloud patches and compares test objects against location-specific memory banks in a shared coordinate frame to ensure spatially coherent anomaly scoring.

Key results

First application of triplet contrastive learning to 3D anomaly detection
Location-specific memory banks eliminate spatial ambiguity in patch comparison
State-of-the-art mean AUROC of 0.901 on the Real3D-AD benchmark
Lightweight, consumer-GPU-trainable architecture robust to manufacturing tolerances

Why it matters

Provides a scalable, highly accurate solution for robotic quality inspection, enabling manufacturers to reliably detect subtle geometric defects beyond human capabilities.

Abstract

Robotic quality inspection is emerging as a key enabler in intelligent manufacturing, allowing robots to tran- scend human limitations in endurance, consistency, and access to complex structures. By detecting subtle defects with speed and precision, robotic inspection enhances efficiency while elevating production quality. While most existing approaches emphasize 2D image-based surface defect detection, they often overlook geometric defects, which are more prevalent and challenging in industrial inspection. To overcome this gap, we formulate geometric defect detection as anomaly detection in 3D point clouds and propose a novel framework that integrates contrastive learning with spatially aware comparisons of local geometries. Specifically, we partition point cloud surfaces into patches and employ contrastive learning to train a neural network-based feature extractor capable of capturing rich geometric representations. An anomaly detection algorithm is then introduced to identify defects by comparing patch-level features in a spatially consistent manner. Evaluated on the recent Real3D-AD benchmark, our method achieves a mean area under the ROC curve of 0.901, establishing a new state of the art and demonstrating the potential of robotic inspection systems to move beyond human limitations in detecting subtle geometric anomalies.

Index terms

Computer Vision for Manufacturing Industrial Robots Factory Automation