Capturing, Generating and Interacting with Virtual Environments of Smart and Flexible Industrial Plants using Language Models

BIRAZ aims to revolutionise industrial Virtual Worlds (VWs), transforming them from static, manually designed environments into dynamic, intelligent ecosystems capable of reasoning. By integrating Large Language Models (LLMs) and their multimodal variants, BIRAZ seeks to optimise automation in the creation of these environments and refine human-machine interaction, enabling more flexible and efficient management of industrial plants, including in Basque.

The adoption of Virtual Worlds and Digital Twins has demonstrated productivity increases of up to 20% and significant reductions in development cycles. However, critical technical barriers remain:
•    Complexity of creation: reliance on manual processes and specialist knowledge.
•    Rigid interaction: interfaces limited to conventional peripherals (keyboard/mouse).
•    Language gap: lack of robust support in languages such as Basque for complex technical interactions.

BIRAZ addresses these challenges through four key areas of research:
1)    Semantic Capture and Reconstruction: moving away from reconstruction based on images, CAD and purely statistical point clouds towards models with semantic awareness. 
2)    Generation and Operational Adaptation: democratising the creation and adaptation of 3D industrial plants for non-specialist staff (line managers or operators).
3)    Multimodal Conversational Interaction: evolving towards natural language and hands-free interfaces.
4)    Basque in Industry 5.0: promoting the use of Basque in the industrial digital twin domain.

BIRAZ will position Basque companies at the forefront of Industry 5.0, enabling industrial plants not only to be digitally represented, but also to be ‘communicated’ and ‘reconfigured’ using human language, breaking down the barriers between the physical and digital worlds in Basque as well.