.png)
Researchers recently released LegoGPT, a novel artificial intelligence (AI) model capable of creating three-dimensional (3D) Lego construction designs. The new AI model is an open-source project that aims to establish if AI models can produce structures that are both compatible with real-world physics and stable. The researchers have released information on how the model was created and have made the dataset available to the public. Humans and robots evaluated the AI-generated Lego buildings to check their stability.
The LegoGPT AI model was built using LLaMA-3.2-Instruct
In a blog post, Carnegie Mellon University researchers described the LegoGPT AI model. The large language model (LLM) can create a Lego building from a text prompt, guaranteeing that it is physically stable and constructible. The open-source model is available for download and usage on GitHub under the permissive MIT licence.
Users may ask the model to create a "streamline elongated vessel" or a "backless bench with armrest," and it will provide a design that not only meets the description but can also be positioned upright without collapsing.
This is achievable because LegoGPT consists of two components: the fundamental AI model and a stability analysis system. The researchers built the basis model using a fine-tuned version of the Llama-3.2-Instruct with one billion parameters. This was combined with Gurobi, a mathematical optimisation solver that does stability analysis on each created structure.
Along with developing the revised architecture, the researchers developed a dataset for training the model on Lego creations. StableText2Lego is a dataset that includes approximately 47,000 Lego constructions and 28,000 distinct 3D objects. Each structure includes full descriptions, design code, and models.
To ensure that the created structures are sturdy, the researchers tested them using a dual robot assembly. The assembly was entrusted with reproducing the drawings and ensuring that they could stand upright. Humans rebuilt some of the designs to examine how they affected stability with less dexterous hands. The study indicates that 99.8 percent of all constructions passed the stability test.
#legogpt #ai #legostructures #legogptaimodel
