World ModelsarXiv preprint2025
Matrix-Game: Interactive World Foundation Model
Jiuming Liu, Chaojun Ni, Mengmeng Liu, Chensheng Peng, Fangjinhua Wang, Sitian Shen, Marc Pollefeys, Masayoshi Tomizuka, Ayush Tewari, Per Ola Kristensson et al.Tsinghua University, BAAI
摘要
We present Matrix-Game, an interactive world foundation model that enables rich interaction with simulated environments. Matrix-Game learns a generalizable world representation that supports dynamic object manipulation, physics-based interactions, and multi-agent scenarios. The model serves as a foundation for training downstream robotic policies and evaluating embodied AI systems in diverse interactive settings.
world foundation modelinteractiveMatrix-Gamesimulationembodied AI
