Overview
Field: Computer Vision Authors: Ava Pun, Kevin Dai, Sally Chen, Maxwell Jones, Ruixuan Liu, Kangle Deng, Changliu Liu, Maneesh Agrawala, Deva Ramanan, Jun-Yan Zhu Published: 2026-10-22 arXiv: 2610.31651
Summary
Frontier coding agents are becoming increasingly adept at 3D modeling. Yet, they still face challenges when building objects meant to be assembled in the physical world, often creating invalid assemblies with floating, colliding, or disconnected parts. The authors argue that providing agents with tools for assembly construction and physical feasibility verification is sufficient to enable them to generate assemblies that are both physically valid and faithful to the input prompt.
Focusing on the LEGO brick domain, the authors present BrickWright, a framework for generating physically feasible LEGO brick builds from text or image prompts using pre-trained agents.
BrickWright guides agents to create builds by writing code with BrickLayer, a Python toolkit with two main components:
- An ergonomic brick assembly API, which reduces reasoning cost by half compared to raw LDraw
- A validator that provides feedback on physical feasibility
Results
Whereas baseline agents achieve physical validity rates ranging from only 9% (GPT-6 Luna) to 69% (GPT-6 Astra), BrickWright-guided agents achieve 100% validity on 512 prompts. In addition, BrickWright improves prompt adherence on lightweight agents while maintaining prompt adherence on stronger ones.
*Auto-collected on 2026-10-24.*