Reproduce “Generating Critic-Guided Synthetic Experience in Latent World Models” 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
Jorge de Freitas, Christos Ziakas, Alessandra Russo
CGSE uses an offline critic and learned world model to generate synthetic transitions for robot manipulation policies. It refines candidate actions, scores them with latent-model lookahead, and stores only a selected action and one-step predicted successor. Across three OGBench tasks, average success improves 7.1% relative to offline FQL, mainly on cube-single; matched controls do not establish added benefit from model-based lookahead.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.