Reproduce “Skill2Real: Agentic Skill Learning for Zero-Shot Sim-to-Real Robot Manipulation” 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
Xincheng He, Siyu Ma, Chang Yu, Yunuo Chen, Yanjia Huang, Ying Nian Wu, Yin Yang, Chenfanfu Jiang
Skill2Real learns reusable manipulation programs in simulation through a shared robot API, then deploys frozen skill memories using public observations. Its Cerebellum learns local skills and its Brain composes them into tasks. On held-out LIBERO-Pro Long tasks, Overall success reaches 56.3%, but position-perturbation success is 36.5%, below instruction-perturbation success of 76.0%; deployment requires robot-specific perception, calibration, and control.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.