Reproduce “RoboCook: Long-Horizon Elasto-Plastic Object Manipulation with Diverse Tools” 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
Haochen Shi, Huazhe Xu, Samuel Clarke, Yunzhu Li, Jiajun Wu
A robotic system named RoboCook enables long-horizon manipulation of deformable objects using diverse tools by integrating data-efficient Graph Neural Networks with self-supervised policy learning. The system successfully completes complex tasks like dumpling making and alphabet cookie shaping, outperforming baselines in shape accuracy and achieving 9.3-second planning steps.
Key insights
- Complex, multi-tool elasto-plastic object dynamics can be accurately learned from sparse visual observations using GNNs with significantly reduced real-world data collection.
- Autonomous tool selection for long-horizon tasks is achievable by combining a point-cloud-based tool classifier with learned dynamics models, enabling flexible and adaptive tool use.
- Self-supervised policy learning, leveraging synthetic data from a learned dynamics model, drastically improves planning efficiency and performance, enabling rapid action synthesis without extensive human demonstrations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.