Reproduce “RH20T: A Comprehensive Robotic Dataset for Learning Diverse Skills in One-Shot” 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
Hao-Shu Fang, Hongjie Fang, Zhenyu Tang, Jirong Liu, Chenxi Wang, Junbo Wang, Haoyi Zhu, Cewu Lu
A comprehensive real-world robotic manipulation dataset, RH20T, comprising over 110,000 multi-modal sequences across 147 diverse tasks, has been collected by researchers at Shanghai Jiao Tong University. This dataset enables robotic policies to achieve improved few-shot learning and generalization capabilities compared to non-pretrained models.
Key insights
- Collecting real-world, large-scale, and multi-modal robotic manipulation data, particularly for contact-rich tasks, requires an advanced and intuitive tele-operation system.
- Exposure to a wide variety of tasks, robot embodiments, and environmental conditions during pre-training is critical for enabling few-shot learning and generalization in robotic policies.
- Integrating diverse sensory inputs (visual, tactile, force, audio, proprioception) allows for learning more robust and versatile manipulation skills, especially for complex interactions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.