Reproduce “Discovering Minimal Reinforcement Learning Environments” 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
Jarek Liesen, Chris Lu, Andrei Lupu, Jakob N. Foerster, Henning Sprekeler, Robert T. Lange
The authors learn a neural contextual bandit environment where reinforcement learning agents practice before evaluation in the original simulator. It supplies a state and immediate reward, then ends after one action; returns there guide updates. On Brax tasks, agents reached competitive returns after 10,000 synthetic steps versus millions typically needed directly, a step-count comparison that excludes costly discovery.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.