#AI #Robotics #GenerativeAI #RobotAssembly #GPT #BloxNet #AI #DesignForAssembly #VLM #AIInConstruction #UCberkeley #ICRA2025 #AIInnovation @Ken_Goldberg @Berkeley_AI 6/6

Remarkably, Blox-Net can generate and assemble designs like giraffes, lighthouses, and tables with minimal human intervention. Five of six structures were assembled successfully 10 times in a row without human intervention. See more at https://t.co/0tbVxVkFmv 5/n

The system prompts GPT to generate designs based on a prompt and available blocks, refines them through physics simulations, and evaluates them with a real robot arm using a suction gripper. It has a top-1 design accuracy of 63.5% and 99.2% block placement accuracy. 4/n

Blox-Net addresses a class of GDfRA problems where the components are rectangular and cylindrical blocks. Blox-Net works in 3 Phases: 1) generating designs using a vision-language model (VLM), 2) simulation for physical constructability, and 3) robot execution. 3/n

The field of "Design for Assembly" has a long history. In a new paper https://t.co/vf0Mu3WjtP (under review for @ieee_ras_icra 2025), we propose a novel class of "Generative Design for Robot Assembly (GDfRA)" problems that use Generative AI and robot motion planning. 2/n

AI can generate text and images, can it generate and construct physical 3D structures? Blox-Net from @UCBerkeley @AUTOLab_Cal takes as input a set of available blocks and a word like "giraffe" and uses ChatGPT and custom software to guide a real 🤖robot to build it. 1/n