Reproduce “UniIntervene++: An Adaptive Intervention Agent for Efficient Real-World Reinforcement 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
Yudong Lin, Haoyuan Deng, Zhuoxuan Yuan, Zaijia Yang, Yuanjiang Xue, Ziwei Wang
UniIntervene++ learns online to allocate robot control among its reinforcement-learning policy and two automated helpers, using scene observations and recent outcomes; a probe periodically tests the policy without assistance. Across five real-world tasks, it averages 89.67% success, versus 83.67% for the best baselines, while human-controlled training steps average 0.77%; overall intervention remains 56.04%.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.