Nvidia, alongside a diverse group of researchers from Stanford, Caltech, and other institutions, has introduced NitroGen, a groundbreaking AI model designed to play a wide array of video games.In a recent LinkedIn post, Jim Fan, Nvidia's Director of AI and Distinguished Scientist, described NitroGen as "an open-source foundation model trained to play 1000+ games." The implications of this technology extend beyond gaming, offering significant benefits for simulations and robotics as well.
Source:
tomshardware.comNitroGen represents an attempt to create a "GPT for actions," applying large-scale training techniques previously used in language and computer vision to the realm of gaming.The research highlights the ambition to develop "generally capable embodied agents that can operate in unknown environments," a long-sought goal in AI research.The foundation of NitroGen is based on the GROOT N1.5 architecture, originally designed for robotics, which suggests that advancements in gaming AI could also enhance robotic capabilities in unpredictable environments.
Sources:
tomshardware.comcoolbyte.irThe AI has been adapted to handle games with diverse mechanics and physics, showcasing its versatility.Researchers utilized over 40,000 hours of public gameplay videos shared by streamers, which included real-time gamepad interactions.This extensive dataset was crucial for training NitroGen to navigate various gaming scenarios effectively.In tests, NitroGen demonstrated success across a wide range of genres, including RPGs, platformers, battle royales, and racing games, indicating its broad applicability in the gaming landscape.
Sources:
tomshardware.comcoolbyte.irWhile the initial results are promising, Fan emphasizes that this is merely the beginning, with significant challenges remaining.The first version of NitroGen focuses on fast motor control, or what Fan refers to as "gamer instinct." The model has shown a "strong competence across diverse domains," achieving a 52% relative improvement in task success rates compared to models trained from scratch, even in procedurally generated worlds and unseen games.
Source:
tomshardware.comThe open-source nature of NitroGen allows enthusiasts in gaming, robotics, and AI to experiment with the model.Researchers have made available the pretrained model weights, the entire action dataset, and the code, encouraging innovation and exploration within the community.This collaborative approach aims to foster further advancements in both gaming and robotics, as developers can build upon the foundational work done by Nvidia and its partners.
Sources:
tomshardware.comcoolbyte.irIn summary, Nvidia's NitroGen is a significant step forward in the development of generalist AI for video gaming.Its ability to adapt to various game types and mechanics not only enhances the gaming experience but also opens new avenues for research and application in robotics and simulations.As the technology evolves, it may lead to more sophisticated AI systems capable of operating in complex, real-world environments, further blurring the lines between virtual and physical interactions in gaming and beyond.