Researchers also enhance the system’s capabilities by adding structure and color options. For example, using the appearance indicators such as “Electric Guitar in Metal Purple”, Leggat can produce a guitar model, in which bricks are assigned purple color.
Testing with robots and humans
To prove their designs in real life, researchers collected the robot to the AI -created Lego models. He used a double robot arm system with a force sensor to take and keep bricks according to AI infield instructions.
Human testers also made some hand designs, which shows that AI makes a real constructive model. The team noted in its dissertation, “Our experiences show that the Leggerpat develops stable, diverse and aesthetically pleasant Lego design that is closely linked to the input text indicators.”
When the 3D creation is checked against another AI system, the league focuses on structural integrity. The team experienced against multiple alternatives, including Lilama mesh And the other 3D Generation model, and its approach produced high percentage of stable structure.
A two robot weapons video provided by researchers is creating a league.
Still, there are some limits. The current version of the league works in just 20 × 20 × 20 building space and uses only eight standard brick types. The team acknowledged, “Our method supports a fixed set of Lego bricks, currently commonly used.” “In the future work, we plan to expand the brick library to include different types of dimensions and brick, such as slopes and tiles.”
Researchers also hope to increase their training data to add more than 21 categories available at this time. Meanwhile, other people can literally build on their work. Researchers released their data, code and models on them. Project website And Guar.