Day two of NVIDIA GTC is behind us. The spotlight of the day was NVIDIA CEO Jensen Huang’s keynote session on the San Jose Enviornment. The sector was packed to the rafters with an estimated 16,000 attendees. Huang offered for almost two hours, and at one level, introduced the Disney BDX robotic up on stage with him.
The Disney BDX robotic entertained the gang whereas NVIDIA CEO Jensen Huang mentioned new robotic merchandise. | Credit score: The Robotic Report
Whereas the session did cowl some robotics information, the principle focus of Huang’s presentation was to persuade the mainstream media that the Deepseek launch final 12 months was truly an incredible factor for NVIDIA’s enterprise. NVIDIA is all in on what it defines as “AI Factories”, the following iteration of knowledge facilities optimized for operating AI workflows primarily based on NVIDIA structure.
This a part of the keynote highlighted the brand new Blackwell chip and new “Dynamo” software program designed to make AI reasoning 40x quicker than earlier generations.
Robotics highlights from the keynote
Huang made a number of important robotics bulletins through the speak, beginning with NVIDIA’s partnership with GM to develop next-gen automobiles, automotive factories, and robots.
Moreover, NVIDIA introduced it’s open-sourcing GR00T N1. Later within the day, I had the chance to take a seat down with Akhil Docca, senior product advertising and marketing supervisor of NVIDIA Omniverse, to debate the importance of this announcement.
NVIDIA designed GR00T N1 for versatility, enabling a single basis mannequin to energy numerous humanoid robots. Builders can fine-tune it for particular purposes and make the most of its artificial knowledge technology capabilities to reinforce coaching datasets. By open-sourcing the mannequin on platforms like Hugging Face, accessibility is elevated, decreasing the barrier for researchers and builders.
The mannequin advances AI robotics by transferring in direction of a generalist expertise able to notion, reasoning, and motion throughout varied situations. This method goals to speed up innovation by offering a single, adaptable AI framework that may carry out a number of duties, selling the event of extra versatile and clever robots.
Builders can now fine-tune the mannequin for particular use instances, making it extra adaptable to totally different robotic purposes. The mannequin comes with a blueprint for producing artificial knowledge, which helps broaden the range of coaching datasets and enhance robotic efficiency throughout varied environments.
NVIDIA introduced Cosmos at CES earlier this 12 months, however at GTC, Huang reiterated the significance of Cosmos for robotic growth. NVIDIA Cosmos is a platform of state-of-the-art generative world basis fashions (WFMs), superior tokenizers, guardrails, and an accelerated knowledge processing and curation pipeline. It’s constructed to energy world mannequin coaching and speed up bodily AI growth for autonomous automobiles (AVs) and robots.
Huang additionally introduced Newton, an open-source, extensible physics engine being developed by NVIDIA, Google DeepMind, and Disney Analysis to advance robotic studying and growth. That is why Huang invited the Disney BDX robotic onstage with him.
NVIDIA Newton diagram. | Credit score: NVIDIA
Newton is suitable with MuJuCo, a extensively used open-source physics engine for robotics analysis and growth, particularly involving advanced dynamics and contact-rich environments.
On the Cosmos and Newton bulletins, Docca mentioned, ”The genesis of that was actually to construct this subsequent technology open supply physics engine for humanoid robots. The concept was to have the ability to herald physics which can be relevant to humanoid robots as inertial movement, and friction, to higher predict how the robotic traverses. How does it deal with dexterous manipulation.”
Agility Robotics and Boston Dynamics announce new NVIDIA enhancements
Yesterday, Agility Robotics introduced firms are also collaborating to make Digit fashions accessible to companions by means of “Mega”, an NVIDIA Omniverse Blueprint.
Utilizing foundational fashions is a recreation changer by way of the rapidity and generality of bringing multi-purpose platforms like Digit into the world. Some highlights of right this moment’s information embrace:
- Agility has an NVIDIA accelerator (GPU) in Digit that runs onboard AI fashions for notion and controls
- Agility makes use of IsaacSim and IsaacLab to coach its AI fashions offline within the cloud
- Agility is engaged on help for the lately introduced NVIDIA Mega platform (a large-scale simulator constructed on high of IsaacSim) in order that prospects like Schaeffler can put Digit into their facility-scale digital twins
In a separate announcement, Boston Dynamics revealed it has expanded its collaboration with NVIDIA to construct the following technology of AI capabilities for humanoid robots. As an early adopter of the NVIDIA Isaac GR00T platform, Boston Dynamics’s Atlas robotic is one among many humanoids utilizing the NVIDIA Jetson Thor computing platform.
The compact dimension, excessive efficiency, and effectivity of Jetson Thor allows Atlas to run advanced, multimodal AI fashions that work seamlessly with Boston Dynamics’ whole-body and manipulation controllers.
Builders at Boston Dynamics and its analysis companions are additionally making speedy breakthroughs in discovered dexterity and locomotion AI insurance policies utilizing Isaac Lab, an open-source, modular framework for robotic studying in bodily correct digital environments, which is constructed on NVIDIA Isaac Sim and NVIDIA Omniverse applied sciences. The 2 firms are collaborating to outline key platform parameters together with practical security and safety architectures, in addition to key studying and pc imaginative and prescient pipelines utilizing NVIDIA’s coaching and simulation platforms.
Dr. Ken Goldberg entertains together with his session
Dr. Ken Goldberg was surrounded by followers with questions following his GTC 2025 session. | Credit score: The Robotic Report
One of many periods I attended within the afternoon was Dr. Ken Goldberg’s “Igniting the Actual Robotic Revolution Requires Closing the Knowledge Hole.” The session was pure Goldberg, and it was standing room solely within the room, as Goldberg took us by means of a historical past lesson on the event of Dexnet, the genesis of Ambi Robotics, and the tempo of innovation for AI over the past 5 years. This can be a session price replaying, as soon as the video is public. Goldberg was additionally a visitor on our 2025 predictions webinar again in January 2025, watch the replay right here.
NVIDIA takes over town
NVIDIA CEO Jensen Huang is on stage with Hewlett Packard Enterprise CEO Antonio Neri on the GTC after-party on South Market Road in San Jose. | Credit score: The Robotic Report
Final 12 months marked the return of a bodily occasion for GTC, following the pandemic. This 12 months, the occasion has grown, spilling out of the San Jose Conference Middle (SJCC) and filling the streets for blocks round SJCC. The occasion fully blocks and gated the realm on South Market between First Road and Almaden Blvd. At lunch, the realm is stuffed with meals vans to feed the 20,000+ attendees, and at evening there’s an evening market with meals and craft distributors and a stage with reside music to entertain the gang.