Reproduce “SimScale: Agentic Simulation Data Scaling for Sim-to-Real Policy Learning” 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
Tianyuan Dai, Kevin Lin, Yunfan Jiang, Minghuan Liu, Junhong Xu, Justin Tran, Benjamin Nguyen, Yingrui Huang and 3 more
SimScale combines a parameterized object-category generator with adaptation of one human demonstration across geometries to train RGB visuomotor policies. AssetEngine creates assets, while DemoEngine adapts behavior across shapes and DexMimicGen expands it across poses. In three bimanual tasks, policies achieved higher average real-world task scores than baselines using one human demonstration; the result applies to evaluated tasks and conditions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.