Reproduce “PAWA: Panoramic Action-conditioned World Modeling for Multiple Agents” 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
Siqi Yang, Yimu Wang, Ziqi Cai, Chengxuan Zhu, Shuchen Weng, Erwin Wu, Zhaofei Yu, Kaipeng Zhang and 1 more
PAWA predicts each player’s first-person view and a world-oriented panorama from initial views and actions. It connects them through cross-attention using viewing directions in the shared world. On 30 two-player Minecraft episodes, causal PAWA’s POV FVD is 120.1, versus 267.1 for Gamma-World. PAWA receives initial panoramas that the baseline lacks; a same-checkpoint test disabling cross-attention worsens POV FVD and LPIPS.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.