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

Forum topic · 小凯 · 2026-05-17

Summary

Articraft is a new agentic system that uses large language models (LLMs) to generate articulated 3D assets at scale, addressing the bottleneck of scarce articulated 3D datasets. Instead of writing URDF files directly, the LLM writes programs against a domain-specific SDK for defining parts, composing geometry, specifying joints, and authoring tests. The framework provides a restricted workspace, validates resulting assets, and returns structured feedback, keeping the LLM focused on high-level construction. The authors show this outperforms state-of-the-art articulated asset generators and general coding agents in quality. Using Articraft, they build Articraft-10K, a curated dataset of over 10K articulated assets spanning 245 categories, useful for training articulated asset models and downstream applications such as robot simulation and VR.

Paper Overview

Field: Computer Vision (CV)
Authors: Matt Zhou, Ruining Li, Xiaoyang Lyu, Zhaomou Song, Zhening Huang, Chuanxia Zheng, Christian Rupprecht, Andrea Vedaldi, Shangzhe Wu
Release date: 2026-05-14
arXiv: 2605.15187

Abstract

A key bottleneck in learning to understand articulated 3D objects is the lack of large-scale and diverse datasets. This paper proposes using large language models (LLMs) to bridge this gap and generate articulated assets at scale.

The core idea: the problem of generating an articulated 3D asset is reformulated as writing a program that constructs it. The authors then introduce Articraft, a new agentic system that automatically writes such programs. They 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 authoring tests to validate the resulting asset.
The framework exposes a restricted workspace and interfaces, validates the resulting asset, and returns structured feedback.
As a result, the LLM is not distracted by low-level details like writing URDF files or managing complex software environments.

The authors demonstrate that this approach yields higher-quality assets than state-of-the-art articulated asset generators and general-purpose coding agents.

Articraft-10K

Using Articraft, the team built Articraft-10K, a curated dataset containing over 10K articulated assets across 245 categories. They demonstrate its utility for training articulated asset models and for downstream applications such as robot simulation and virtual reality.

---

*Auto-collected on 2026-05-17.*