Articraft: An Agentic System for Scalable Articulated 3D Asset Generation

Field: Computer Vision Authors: Matt Zhou, Ruining Li, Xiaoyang Lyu, Zhaomou Song, Zhening Huang, Chuanxia Zheng, Christian Rupprecht, Andrea Vedaldi, Shangzhe Wu arXiv: 2605.15187

Abstract

Learning to understand articulated 3D objects is bottlenecked by the lack of large-scale and diverse datasets. In this paper, we propose leveraging Large Language Models (LLMs) to bridge this gap and generate articulated assets at scale. We reduce the problem of generating articulated 3D assets to writing a program that constructs it. We then introduce a novel agentic system, Articraft, that automatically writes such programs. We design a programming interface and framework that help the LLM do this effectively. The LLM writes code against a domain-specific SDK for defining parts, composing geometry, specifying joints, and writing tests to validate the resulting asset. The framework exposes restricted workspaces and interfaces to the LLM, validates the resulting asset, and returns structured feedback. This way, the LLM is not distracted by details such as writing URDF files or managing complex software environments. We show that this produces higher-quality assets than state-of-the-art articulated asset generators and general coding agents. Using Articraft, we construct Articraft-10K, a curated dataset containing over 10K articulated assets spanning 245 categories, and demonstrate its utility in training articulated asset models as well as downstream applications such as robot simulation and virtual reality.

*Automatically collected on 2026-05-17.*