Case Study: KITECH achieves a synthetic data pipeline for manufacturing AI with Unity

Korea Institute of Industrial Technology builds a digital twin in 14 weeks with Unity

The Korea Institute of Industrial Technology (KITECH) faced the challenge of creating a digital twin for a metal casting foundry without CAD source files and generating scarce, high-quality training data for human-robot collaboration (HRC) scenarios, which is difficult and costly to capture in the real world due to safety and labeling constraints. They partnered with Unity to utilize its Unity Industry suite of tools, including Unity AI, Asset Manager, Asset Transformer, and Version Control, to address this.

Unity implemented a solution that began with converting point cloud data into a simulation-ready digital twin. They used Unity AI to accelerate development, optimize assets, and automate the creation of UI and visual effects. The digital twin was then grounded against real-world measurements and calibrated to minimize the real-to-sim gap, transforming it into a synthetic data pipeline that automatically generates accurately labeled training data. The results were significant: over 100,000 raw frames of synchronized sensor data were captured from 17 dataset episodes, reducing manual labeling effort and providing a scalable source of crucial HRC training data for manufacturing AI.

Korea Institute of Industrial Technology (kitech)