Reproduce “VeriFine: Scaling Verification for Self-Improvement in Embodied Reasoning” 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
Zewei Zhou, Rachel Luo, Yulong Cao, Chaowei Xiao, Chensheng Peng, Boyi Li, Thomas Tian, Zheng Lian and 5 more
VeriFine adapts a policy, its curriculum, and its judge as failures change. Its reference-free rubric judge evaluates reasoning against context, while policy and judge loops select examples and have humans refine unreliable judgments. Across three driving rounds, reasoning scores rise from 60.56 to 71.61, an 18.2% relative gain; evidence covers driving and robot navigation, not every task.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.