Reproduce “Game2Player: Scaling Gameplay Experience for Visual Agents through Environment-Player Co-Development” 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

Xirui Jiang, Yuxi Chen, Rui Yang, Feng Luo, Pengcheng Wang, Bingxuan Li, Zihao Zhao, Yushan Li and 6 more

Game2Player adapts existing browser games into tasks that players can test and report on, with the Builder checking reports against recorded play. It records code-policy and screenshot-based play, then fine-tunes visual agents with sampling weighted toward measured progress. On PlayerBench, Game2Player-9B raises progress from 14.3% to 31.2%; evaluations use one run per task and pause games during decisions.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.