NITROGEN is trained on 40,000 hours of video gameplay
NitroGen just won CVPR Best Paper Honorable Mention!! We are making strides towards general-purpose embodied agents that master not only the real world physics, but also all possible physics across a multiverse of simulations.
It’s been 4 years since MineDojo, our first embodied agent in Minecraft, won NeurIPS Best Paper. Congrats to everyone on the team!!
Many users congratulated the NitroGen team on its CVPR Best Paper Honorable Mention because they see the work on generalist gaming agents as exciting and impactful.
Check out the NVIDIA blog!
@DrJimFan @guanzhi_wang @loicmagne_ Congrats to the team 💚
This is a major step.
General-purpose embodied agents trained across many simulated worlds will not only learn tasks.
They will learn how to adapt to changing rules of reality.
That is powerful.
But it also changes the safety question.
If an agent masters many environments, physics models and reward structures, we need to observe not only what it can do, but whether its trajectory remains aligned with the original human intent when the environment changes.
Simulation teaches capability.
But deployment needs trajectory observability.
Because the real risk is not only failure.
It is a capable agent adapting successfully while the goal quietly shifts.
@DrJimFan @guanzhi_wang @loicmagne_ Congrats to the team!!
@georgiagkioxari @guanzhi_wang @loicmagne_ Thanks Georgia!! It’s such a great privilege to coauthor with you!
@DrJimFan @guanzhi_wang @loicmagne_ Congrats to the team!!
@DrJimFan @guanzhi_wang @loicmagne_ lol I tried asking the MineDojo assistant how to prep for class once and it told me to mine 8 diamonds first
physics across all simulations is cool and all but does it know interior design
@jankautz @NVIDIAAI Congratulations Jan!
@NVIDIAAI @guanzhi_wang @loicmagne_ 💚💚💚 go team green!
@DrJimFan @guanzhi_wang @loicmagne_ Big congrats Jim!
@DrJimFan @guanzhi_wang @loicmagne_ Congrats Jim (also @guanzhi_wang) !!
@DrJimFan @guanzhi_wang @loicmagne_ "All possible physics"- level is impressively top breakthrough!
@DrJimFan @guanzhi_wang @loicmagne_ Congrats and well deserved, Jim!
@DrJimFan @guanzhi_wang @loicmagne_ Congratulations. Embodied AI is moving from “learn this game” to “learn the grammar of worlds”.
The prize is not one simulator mastered. It is transfer across physics, interfaces and rulebooks without the agent losing its hands.
@DrJimFan @guanzhi_wang @loicmagne_ Congrats!! Really much fun reading from MineDojo to NitroGen, the insights are also applicable into robotics :)
@DrJimFan @guanzhi_wang @loicmagne_ Congratulations!
@DrJimFan @guanzhi_wang @loicmagne_ Huge congrats! Embodied agents tackling multiverse physics feels like the next frontier.
@DrJimFan @guanzhi_wang @loicmagne_ this is a remarkable achievement. given the complexity and ambition of their project, i'd love to hear more about how they plan to bridge the gap between simulated physics and the real world.
@jankautz @NVIDIAAI Congrats!
@DrJimFan @guanzhi_wang @loicmagne_ Wow congratulations Jim and the team!
@DrJimFan @guanzhi_wang @loicmagne_ 4 years from one minecraft to all of them