U.S. Navy photo by John F. Williams · PUBLIC DOMAIN · Wikimedia Commons · Source
At GTC Berlin on October 21, 2026, NVIDIA officially announced a major expansion of its ‘NVIDIA Isaac’ platform, designed to support the development of general-purpose humanoid robots. By unifying robot development pipelines that were previously fragmented from simulation to real-world deployment and providing next-generation virtual datasets and reference models, the initiative is expected to significantly accelerate the commercialization of physical AI-based robotics.
Isaac Platform Expansion at GTC Berlin: Building a Physical AI Ecosystem
During his keynote at ‘GTC Berlin’ on October 21, 2026, NVIDIA founder and CEO Jensen Huang announced the massive expansion of the ‘NVIDIA Isaac’ general-purpose robotics platform. Emphasizing that a ‘ChatGPT moment’ for robotics is fast approaching, Huang stated that the company would provide models, data, and simulation tools across the industry via a full-stack open robotics platform. The announcement focused on providing an integrated development environment that links everything from data collection for robot intelligence training to physical robot control.
Reference Humanoid ‘Agile ONE SN’: Hardware Specs Powered by Jetson Thor and GR00T
In conjunction with the platform expansion, NVIDIA collaborated with global robotics company Agile Robots to unveil the full-stack reference humanoid, ‘Agile ONE SN.’ The Agile ONE SN stands 1.40m tall, weighs 40kg, and offers a total of 31 degrees of freedom (DoF). Its arms can each handle a payload of approximately 3kg, and its hands feature the ‘WUJI Hand 2,’ a 5-fingered precision hand with 20 DoF, 1kHz real-time force control, and tactile sensors for intricate object manipulation.
The Agile ONE SN’s computing and sensor array is powered by the NVIDIA ‘Jetson Thor’ onboard computing platform, which provides 2,070 teraflops (TFLOPS) of performance at FP4 and 128GB of unified memory, enabling real-time inference and full-body control on the robot itself. It perceives its surroundings via multimodal sensors including head and wrist-mounted cameras, LiDAR, and Inertial Measurement Units (IMU). By integrating virtual environment simulation with real-world robot operation into a single platform, the time required for system setup—previously lasting several weeks—has been drastically reduced to just a few days.
Isaac Lab 3.0 and Isaac OS: Reducing Runtime from Simulation to Deployment
NVIDIA also announced advancements in new runtimes and software framework updates. The now officially released ‘NVIDIA Isaac Lab 3.0’ is an open-source robot learning framework that leverages multi-GPU workflows, from desktops to data centers, to train robot motion policies. The platform integrates support for NVIDIA PhysX and the new ‘Newton’ physics engine as simulation backends.
Alongside this, ‘NVIDIA Isaac OS’ was launched, an open-source runtime framework based on NVIDIA Holoscan. This system links AI models, sensor data, control algorithms, and robot hardware in a deterministic structure in real-time, ensuring that control policies generated in virtual simulations can operate on physical robots immediately and without distortion.
Massive Synthetic Datasets and the ‘Halos’ Safety Platform
To build a virtual data ecosystem essential for robot learning, the ‘SimReady Central Library,’ which provides over 2,000 assets for simulation, and the ‘Human 1M Dataset,’ containing 3D hand tracking and synchronized video data, were released via Hugging Face. Additionally, an open-source reference workflow based on the NVIDIA DGX Station was introduced to support the generation of high-quality synthetic data using ChatGPT Astra and NVIDIA Cosmos world models.
To enhance the safety of automated facilities and robot operations, the full-stack safety platform ‘NVIDIA Halos,’ based on NVIDIA IGX Thor, was also introduced. The Halos safety stack is designed to prevent accidents by allowing robots to actively respond to dynamic, changing environments rather than remaining confined to fixed virtual safety zones.
Expanding European Robot Ecosystem and Industrial Partnerships
The presentation highlighted widespread adoption of the Isaac platform by global robotics firms, including in Europe. ABB Robotics, KUKA, Toyota Material Handling Europe in the logistics sector, and SoftServe reported that they have reduced robot deployment times by up to 50% by adopting NVIDIA Omniverse and Isaac Sim.
- Agile Robots and Humanoids: Development of the ‘Agile ONE SN’ reference humanoid and integration of Jetson Thor-based control systems.
- Industrial Robotics and Manufacturing: ABB Robotics (RobotStudio HyperReality integration), KUKA (FactoryLens simulation integration), Sereact.
- Autonomous Mobile and Logistics: Toyota Material Handling Europe, Bear Robotics (KR1 perception module), Bellhop (Hotel service robot ‘Billy’).
Release Schedule and Limitations of Single-Source Information
The ‘Agile ONE SN’ reference humanoid is scheduled for official availability in early 2027, with world-renowned research institutions such as Stanford University and ETH Zurich expressing interest in early adoption. NVIDIA expects that by sharing data and learning frameworks on a unified standard platform, both academia and industry will accelerate the proliferation of humanoid technology.
However, the detailed specifications and partnership outcomes in this article are based solely on official NVIDIA press releases. Actual mass-produced product performance, yields, and hardware compatibility at the time of commercialization in 2027 will require independent verification post-release. Furthermore, adherence to safety certification procedures and legal regulations in various industrial settings remains a key challenge to monitor in the future.
Official Source Information:
- NVIDIA Newsroom Press Release: NVIDIA Advances Isaac Platform for General-Purpose Robotics With New Reference Humanoid, Frameworks, Simulation Pipelines and Data (Released October 21, 2026)