Nvidia-led NitroGen is a generalist video gaming AI that can play any title — research also has big implications for robotics

Researchers from Nvidia, Stanford, Caltech, and other institutions have developed NitroGen, an open-source AI model trained to play over 1000 video games. The implications extend beyond gaming to robotics and simulations, aiming to create generally capable agents for unknown environments. NitroGen, based on the GROOT N1.5 architecture, shows promise in diverse game genres and procedural worlds, with a 52% improvement in task success rates over models trained from scratch. The research is open-source, inviting exploration and modifications for those interested in gaming, robotics, and large language models (LLMs).