Reproduce “Code to Control: Synthesizing Parameterized Reactive Controllers” 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
Zergham Ahmed, Joshua B. Tenenbaum, Chris Bates, Samuel J. Gershman
Code to Control learns executable policies mapping structured states directly to actions. A language model revises policies using trajectories and rewards, and the resulting controller acts without language-model calls or planning at decision time. On Pong, median best score was +18 versus −19 for WorldCoder; the experiments use structured states and target reactive control.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.