Industrial

The "Brownfield" Challenge

Physical AI and the Industrial Metaverse work perfectly in new factories (“greenfield”). But the reality is “Brownfield": established structures, old machinery, and missing CAD data. How do we bring these facilities into the metaverse? 

 

A Comparison of Technologies

In the real-world lab, we evaluated various methods for efficiently digitizing the physical world: 

 

1. Manual modeling: The gold standard for precision (93.4% mAP), but extremely time-consuming. 
2. 3D Gaussian Splatting: The game changer for brownfield sites. A simple video scan delivers near-hand-scanning quality (91.2% mAP) in just minutes. This enables the rapid digitization of entire halls. 
3. Generative AI: Ideal for contextual objects. Tools that generate 3D models from text or images are fast, but they do not yet achieve the industrial precision required for critical parts. 

 

Added Value: Shop Floor Management

A digital twin must be cost-effective. Using technologies such as Gaussian splatting, we map existing facilities cost-effectively to achieve specific goals: 

• Space Management: Will the new machine fit into the existing space? 

• Inventory: Automatic inventory tracking using cameras. 

• Logistics: Optimization of driving routes for autonomous systems. 

Conclusion

The Future of Physical AI is hybrid: Gaussian splatting for the environment, Synthetic data for training robots, AI models, and GenAI for scaling. This finally makes the Industrial Metaverse feasible for existing systems as well.