Engineers at Carnegie Mellon University in the United States have introduced a new artificial intelligence model called LegoGPT. It promises to simplify the process of assembling objects from Lego constructor.

The developers claim that LegoGPT is the first neural network capable of generating Lego assembly instructions taking into account the stability of constructed objects so that they do not fall apart after completion. The model was trained on a database of 47,000 sets. The neural network allows users to receive precise instructions on how to build various pieces, ranging from simple to more complex structures.

However, LegoGPT has its limitations. The neural network functions in a “cube” with dimensions of 20×20 points, which limits the width, height, and length of structures to 20 standard pieces. In addition, it works with 8 basic types of elements - from 1×1 squares to 2×6 rectangles.

The authors of the project plan to expand the library of tools and improve the functionality of the model to create more complex structures. It is expected that in the future LegoGPT may find application in production tasks, including the manufacture of furniture according to individual design