Reproduce “Verifiable Hidden Dynamics Play: Generating Agentic RL Environments from Solved Mechanisms” 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
Xinjie Shen, Wei Fan, Xudong Guo, Jianhong Tu, Yang Su, Chuqiao Kuang, Yinger Zhang, Dayiheng Liu
VHD-Play generates interactive optimization environments by first sampling and solving a mathematical decision problem, then rendering its dynamics as tools while retaining solver-derived references for scoring. Training on three mechanism families raises Qwen3.6-35B-A3B’s five-family macro from 0.204 to 0.815 and improves external storefront performance. This supports feasibility and transfer, not general agentic competence or causal attribution.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.