Artisan: Physically Grounded Articulated 3D Asset Generation with a Multi-Agent Pipeline

Forum topic · 小凯 · 2026-10-13

Summary

Researchers present Artisan, a system for physically grounded articulated 3D asset generation that shifts the focus from reproducing an object's appearance to capturing how it moves and functions. Building on the Articraft method, Artisan introduces a multi-agent pipeline with physical simulation feedback, extending generation beyond geometry, appearance, and plausible articulation to include collision-free motion, stability under gravity, and accurate mass and material properties. The work also introduces a unified benchmark covering four aspects: spatial fidelity, physical plausibility, downstream task performance, and overall quality, comprising 100 reference assets from ArtVIP, 50 stability prompts, and 50 ABO objects with reference masses. Extensive evaluation of more than ten representative models shows Artisan is competitive in spatial fidelity while achieving the lowest collision rate and highest stability rate. Together, the system and benchmark point toward physically credible, functionally usable 3D generative assets. Paper: arXiv 2610.13947.

Paper Overview

Field: Computer Vision (CV) Authors: Xudong Cai, Ruijie Zhu, Haoliang Shang, Ruining Li, Marc Pollefeys, Shangzhe Wu, Andrea Vedaldi, Chuanxia Zheng arXiv: 2610.13947

Abstract

In this paper, we study physically grounded articulated 3D asset generation --- moving the goal from reproducing how objects look to capturing how they move and function. Inspired by the literature on procedural and agentic generation, we present Artisan, a system that extends the latest method Articraft by introducing a multi-agent pipeline with physical simulation feedback, enabling comprehensive attributes beyond geometry, appearance, and physically plausible articulation, such as collision-free motion, stability under gravity, and accurate mass and material properties.

Furthermore, we construct a unified benchmark spanning four key aspects: spatial fidelity, physical plausibility, downstream task performance, and overall quality. The benchmark encompasses 100 reference assets from ArtVIP, 50 stability prompts, and 50 ABO objects with reference mass.

Key Findings

Extensive evaluation over ten+ representative models shows Artisan is competitive in spatial fidelity while achieving the lowest collision rate and highest stability rate.
The system and benchmark together provide a direction for advancing 3D generation toward physically credible, functionally usable assets.

*Auto-collected on 2026-10-13.*