Reproduce “MatchMaker: Automated Asset Generation for Robotic Assembly” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yian Wang, Bingjie Tang, Chuang Gan, Dieter Fox, Kaichun Mo, Yashraj Narang, Iretiayo Akinola
MatchMaker is an automated pipeline that generates diverse, simulation-compatible assembly asset pairs by combining a Vision-Language Model for semantic understanding, a 3D generative diffusion model for shape completion, and a clearance specification module. This approach yielded a higher asset diversity score of 0.43 compared to existing datasets and enabled policies trained on these assets to achieve an 82% success rate in real-world robotic assembly tasks.
Key insights
- Framing paired assembly asset generation as a constrained shape completion problem, guided by extracted contact surfaces, enables automated and diverse asset creation.
- Vision-Language Models can provide high-level semantic reasoning and assembly axis identification to effectively guide lower-level geometric processing for contact surface extraction.
- The integration of a B-rep based 3D generative model with a precise, physics-informed clearance specification step is essential for producing simulation-compatible and physically realizable assembly assets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.