New FORGE dataset reveals that multimodal AI models struggle with manufacturing tasks due to lack of domain knowledge, not visual understanding limitations.

Researchers introduced FORGE, a high-quality multimodal dataset combining 2D images and 3D point clouds with fine-grained manufacturing semantics like exact model numbers

18 state-of-the-art MLLMs were evaluated on three manufacturing tasks: workpiece verification, structural surface inspection, and assembly verification

Analysis shows visual grounding is not the primary bottleneck; instead, insufficient domain-specific knowledge limits performance in manufacturing scenarios

The dataset addresses critical data scarcity and lack of domain semantics in existing evaluations for autonomous manufacturing execution