Evaluating FoundationPose object registration accuracy for AR-guided industrial inspection
Abstract
Augmented Reality (AR) can enhance industrial processes like manual inspections by increasing safety and efficiency. This is achieved by superimposing virtual guidance directly onto the physical environment, such as visual highlights, step-by-step instructions, or animated tool paths. The accurate alignment of virtual guidance depends directly on the precise registration of real-world components. This study evaluates the 6D pose estimation accuracy of the FoundationPose framework for registering real-world, complex industrial components (e.g., fittings, flanges) from a single RGB-D image without object-specific retraining. In a controlled pre-deployment evaluation with a robot-positioned RGB-D camera, the measured accuracy meets the requirements derived for AR-assisted inspection, achieving median translational deviations between 0.5 mm and 2.8 mm and a median rotational error of approximately 1.9° for plane-symmetric components. While performance remained stable against partial occlusions and reflective surfaces, the results highlight that accuracy is limited by key challenges, including performance degradation with symmetric and texture-poor objects and the need for optimization for real-time mobile deployment. Despite these limitations, the baseline accuracy identified in this study indicates the potential of FoundationPose for industrial AR applications that demand precise alignment of virtual overlays, subject to validation on the target device.
Related
- Enhanced discrepancy-aware visual inspection with a single global threshold for industrial assembly
- Accuracy evaluation of railway track irregularity inspection using bogie-mounted IMUs of different grades
- Optimal exposure time for satellite imaging? A trade-off between residual tip-tilt and registration accuracy
- On the accuracy of nonlinear image registration in portable low-field 3D brain MRI
- From Laboratory Accuracy to Industrial Deployment: A Review of Hyperspectral Imaging for Cereal Grain Quality and Safety Assessment
BibTeX
@article{masuhr2026evaluating,
title = {{Evaluating FoundationPose object registration accuracy for AR-guided industrial inspection}},
author = {Christian Masuhr and Julian Koch and Tim Oels and Thorsten Schüppstuhl},
journal = {Scientific Reports},
year = {2026},
month = nov,
doi = {10.1038/s41598-026-74702-1},
url = {https://www.nature.com/articles/s41598-026-74702-1},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Evaluating FoundationPose object registration accuracy for AR-guided industrial inspection
AU - Christian Masuhr
AU - Julian Koch
AU - Tim Oels
AU - Thorsten Schüppstuhl
PY - 2026
DA - 2026/11/06
JO - Scientific Reports
DO - 10.1038/s41598-026-74702-1
UR - https://www.nature.com/articles/s41598-026-74702-1
AB - Augmented Reality (AR) can enhance industrial processes like manual inspections by increasing safety and efficiency. This is achieved by superimposing virtual guidance directly onto the physical environment, such as visual highlights, step-by-step instructions, or animated tool paths. The accurate alignment of virtual guidance depends directly on the precise registration of real-world components. This study evaluates the 6D pose estimation accuracy of the FoundationPose framework for registering real-world, complex industrial components (e.g., fittings, flanges) from a single RGB-D image without object-specific retraining. In a controlled pre-deployment evaluation with a robot-positioned RGB-D camera, the measured accuracy meets the requirements derived for AR-assisted inspection, achieving median translational deviations between 0.5 mm and 2.8 mm and a median rotational error of approximately 1.9° for plane-symmetric components. While performance remained stable against partial occlusions and reflective surfaces, the results highlight that accuracy is limited by key challenges, including performance degradation with symmetric and texture-poor objects and the need for optimization for real-time mobile deployment. Despite these limitations, the baseline accuracy identified in this study indicates the potential of FoundationPose for industrial AR applications that demand precise alignment of virtual overlays, subject to validation on the target device.
SN - 2045-2322
ER -