Reproduce “GAMMA: Generalizable Articulation Modeling and Manipulation for Articulated Objects” 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
Qiaojun Yu, Junbo Wang, Wenhai Liu, Ce Hao, Liu Liu, Lin Shao, Weiming Wang, Cewu Lu
Researchers from Shanghai Jiao Tong University developed GAMMA, a framework that enables robots to model and manipulate diverse articulated objects, including novel categories, from point clouds. The system achieved an 86.86% success rate for pushing a door on unseen categories and reduced articulation axis error to 8.85 degrees compared to baseline methods in similar scenarios.
Key insights
- Combining robust articulation modeling with part-aware grasp pose affordance enhances the actionability and effectiveness of robotic manipulation for diverse objects.
- A physics-guided adaptive manipulation loop, which refines articulation parameters through interaction, significantly improves robustness to initial modeling errors.
- Generalizability to unseen object categories is achieved by learning a comprehensive understanding of kinetic constraints rather than relying on category-specific priors.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.