Reproduce “SUGARv2: Learning Diverse Visual Humanoid Loco-Manipulation Skills from Synthesized Interaction References” 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
Tianshu Wu, Tianxiang Gui, Yue Chen, Tingxu Liu, Ziye Peng, Guangjie Yang, Xiangqi Kong, Ye Shi and 2 more
SUGARv2 adapts reusable task templates to new robot and object layouts, generating coordinated motion references for humanoid loco-manipulation. It trains separate policies for each task using a shared reference-based objective, with teacher supervision continuing during reinforcement learning. Across five simulation tasks, it reports 91.0% mean success versus 39.6% for VIRAL; simulation-trained policies were also deployed on a real robot without fine-tuning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.