Researchers at Carnegie Mellon University’s School of Computer Science have developed a new AI tool that could transform manufacturing and construction.
From Text to Stable Lego Creations
BrickGPT, developed at Carnegie Mellon, uses text prompts to turn ideas into Lego creations. For example, typing “guitar” generates a step-by-step, brick-by-brick guide to building a stable model. While now focused on Lego, the ability to turn words into stable designs has far broader uses.
“This research advances generative manufacturing, enabling anyone to use AI to design and build everyday objects,” said Jun-Yan Zhu. “It’s a new frontier beyond creating videos or photos.” Lego bricks are just the starting point.”
The team says combining AI and robotics could speed up turning ideas into physical form. “By integrating generative AI, we can boost efficiency and remove many of the barriers that slow down prototyping in manufacturing.”
Right now, the BrickGPT demo can create step-by-step instructions for humans or robots to assemble 21 different Lego models, such as a birdhouse, sofa, or piano. To make a sofa, for example, a user types “sofa” into BrickGPT, which generates a 3D model. An algorithm then converts this model into a Lego brick structure, and BrickGPT verifies its stability. The resulting guide can be followed by either a person or a robotic arm to build the final piece.
Creating the StableText2Brick Dataset and Training the Model
To build BrickGPT, researchers made StableText2Brick, a dataset of 47,000 Lego structures from 28,000 voxelized 3D objects with captions, and trained an LLM to predict each brick for stability.
If an error occurs, BrickGPT initiates a rollback process to remove unstable points.“If a structure is unstable, the model reverts to a stable point,” said doctoral student Ava Pun, noting a physics algorithm scores each brick’s stability.
Led by professors Jun-Yan Zhu and Changliu Liu, the team plans to expand BrickGPT beyond 21 objects and diversify pieces to boost accuracy and complexity.
Read the original article on: Techxplore
Read more:Scientists Develop the First Synthetic Tongue that Mimics Human Taste
