Artisan Multi-Agent 3D System Posts Lowest Collision Rates

A multi-agent system called Artisan posts "the lowest collision rates and highest stability rates among the evaluated methods" for generating articulated 3D assets, according to a preprint posted on arXiv by Xudong Cai, Ruijie Zhu, Haoliang Shang, Ruining Li, Marc Pollefeys, Shangzhe Wu, Andrea Vedaldi and Chuanxia Zheng.

The authors frame the task as "moving the goal from reproducing how objects look to capturing how they move and function." Artisan extends an earlier method called Articraft by wrapping it in "a multi-agent pipeline with physical simulation feedback," aimed at "collision-free motion, stability under gravity, and accurate mass and material properties."

Alongside the system, the paper ships a benchmark covering four aspects: spatial fidelity, physical plausibility, downstream task performance and overall quality. It is built on 100 reference assets from ArtVIP and LightWheel, 50 stability prompts, and 50 ABO objects with reference masses, run against more than 10 representative models. The abstract publishes no absolute collision or stability figures, so the "lowest / highest" ordering is the only quantified summary readers get without opening the PDF.

It lands in a week where our robotics tracker has been crowded by funding rounds and warehouse operations stories; a benchmark preprint is the quieter end of the beat.