LegoGPT is an AI that generates real Lego sets by text query

Researchers at Carnegie Mellon University have presented LegoGPT -a unique artificial intelligence model that can create Lego sets based on natural language descriptions. Simply put: you write down what you want to build – and the AI produces step-by-step instructions with a stable construction.

How does LegoGPT work?

Unlike ChatGPT, which predicts the next word, LegoGPT predicts the next brick. The model is trained on a database of over 47,000 constructions, each of which has been annotated with a signature and detailed 3D data. A total of 28,000 unique Lego components were used in the dataset.

The developers say the model creates structures that are 98% physically stable, meaning they can actually be assembled – and they won’t fall apart.

What’s different from other AI builders?”

Autonomous Lego generators have existed before, but LegoGPT differs in its approach to robustness and step-by-step design. Instead of random shapes, the model generates practical and logical structures that follow the laws of physics and structural logic.

“Unlike mere shape generators, our model can think like an engineer,” the authors of the project emphasize.

What can LegoGPT build?”

So far, AI is performing best in furniture objects and simple household items. Demonstration projects include:

• chairs
• sofas
• tables
• bookshelves

The model has already been posted on GitHub, where anyone can check out the architecture, dataset, and results.

Opportunities Without Borders

Although the builds look simple, the potential of LegoGPT is impressive. In the future, such models will be able to:

• create interactive kits for children
• help interior designers visualize ideas
• use in education
• automate the creation of physically correct 3D models

The article LegoGPT – AI that generates real Lego sets by text query was first published on ITZine.ru.