Reproduce “PlayGen2: Learning Real-Time Playable Worlds with Explicit 3D State” 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
Jihai Zhang, Yuhan Zhou, Junyou Li, Deheng Ye, Qi Dou, Yu Cheng
PlayGen2 models persistent 3D world state for playable Doom-like environments, updating local occupancy maps and identity-tracked entities from actions, history, and geometry while a separate renderer generates images. In a controlled pursuit test, it reaches a proximity threshold in 55% of trials within 200 steps. Training used about 500 hours from Freedoom2 MAP01, limiting claims of broader generality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.