The researchers also expanded the system capabilities by adding the touch and colors options. For example, using a appearance wave such as “Electric Guardian in Metal Purple”, LEGOGPT can generate a guitar model, with a brick of purple color.
Test with robots and human beings
To prove that their designs worked in real life, the researchers collected the LEGO models created from AI. They used a dual -robot arm system with power sensors to pick up bricks and put them according to the instructions created by artificial intelligence.
Human laboratories have also built some hand designs, indicating that artificial intelligence creates truly constructive models. “Our experiences show that LEGOGPT produces stable, varied and cosmetic LEGOs that are closely corresponding to the entry text claims,” the team indicated in its paper.
When tested against other artificial intelligence systems to create a three -dimensional, legogpt stands out by focusing on structural integrity. Test the team against many alternatives, including Lama Mish And other three -dimensional generation models, and found that its approach produced the highest percentage of stable structures.
A video clip of two robot weapons that build Lecgpt, presented by researchers.
However, there are some restrictions. The current version of LEGOGPT only works within the construction area of 20 x 20 x 20 and uses only eight standard types of bricks. “Our way currently supports a fixed group of LEGO bricks commonly used,” the team admitted. “In future work, we plan to expand the brick library to include a wide range of dimensions and brick types, such as slopes and tiles.”
The researchers also hope to expand their training data range to include more than 21 categories currently available. Meanwhile, others can literally build their work – researchers released a collection of data, symbol and models on them Project site and Gyrroup.
