NVIDIA Expands Isaac Robotics Platform with Humanoid Reference Platform and New Tools

Phoenix Technology News (Author/ Yu Lei) Beijing Time October 21, NVIDIA announced at the GTC Berlin conference the expansion of the Isaac open robot development platform, adding a humanoid robot reference platform, runtime framework, and dataset, and updating robot learning tools to cover the development process from data generation, simulation training to deployment on physical robots.

NVIDIA founder and CEO Huang Renxun stated that general-purpose robots are rapidly developing, and the 'ChatGPT moment' in the robotics field is accelerating. The tools released this time are intended to help developers integrate models, data, simulation, and computing systems, advancing general-purpose robots into practical applications.

In terms of hardware, Agile Robots has launched the humanoid robot reference platform Agile ONE SN, planned for release in early 2027. The platform comes pre-configured with NVIDIA Isaac GR00T, featuring the Jetson Thor computing platform, primarily targeting robot research and development. According to the release, it integrates simulation, training, and deployment tools into a single system, reducing the initial configuration time from weeks to days.

Agile ONE SN uses the Agile ONE S humanoid robot body, standing at 1.4 meters tall and weighing 40 kilograms, with 31 degrees of freedom, and a single arm payload capacity of 3 kilograms. It is equipped with two WUJI Hand 2 five-finger dexterous hands, each with 20 degrees of freedom, supporting 1kHz force control and tactile perception. The onboard Jetson Thor provides 2070 TFLOPS of FP4 computing power and 128GB unified memory, with sensors including head and wrist cameras, LiDAR, and inertial measurement units.

In terms of safety features, the platform plans to integrate robot-side safety protocols with NVIDIA Halos safety system, though this integration is yet to be completed. NVIDIA stated that institutions such as Stanford University and ETH Zurich have expressed interest in the platform.

In terms of software, Isaac Lab 3.0 is now officially available, supporting multi-GPU training workflows, as well as physics engines such as Newton and PhysX. The newly added open-source runtime framework Isaac OS is built on Holoscan, used to connect AI, sensors, control systems, and robot hardware, supporting deterministic real-time operation from simulation to physical robots.

NVIDIA also released three open-source reference workflows for DGX Station, respectively used for optimizing the GR00T 1.7 model through reinforcement learning, training dexterous manipulation skills, and using ChatGPT Astra to post-train the Cosmos world model to generate synthetic data.

In terms of data, SimReady Central Library provides over 2000 items of robot and factory assets usable for simulation; the Human 1M dataset includes various human activities, synchronized three-view video, and 3D hand tracking data. Both are available on Hugging Face, with the complete Human 1M dataset open to Cosmos Coalition partners.

Besides ABB Robotics, KUKA, and Unitree, several other robotics companies are also adopting NVIDIA-related tools. Among them, ABB Robotics has integrated the Omniverse library into RobotStudio HyperReality, used for simulation and virtual commissioning of robot systems prior to installation; NEURA Robotics utilizes Isaac Sim and Isaac Lab to generate synthetic training data and train robot skills in a simulated environment. Return to the previous page for more information