IT之家 On October 21, NVIDIA announced on October 21 local time at the GTC Berlin conference the expansion of its Isaac open robot development platform, adding a humanoid robot reference platform, a runtime framework, simulation assets and datasets, and updating robot learning tools, covering the development pipeline from training data generation and simulation training to physical robot deployment.
On the hardware side, NVIDIA partnered with Agile Robots to release the humanoid robot reference platform Agile ONE SN, planned for launch in early 2027.
The platform comes pre-configured with NVIDIA Isaac GR00T and equipped with the Jetson Thor computing platform, primarily targeting robotics research and development. By integrating simulation, training and deployment tools, Agile Robots said the platform can reduce upfront environment setup time from weeks to days.
NVIDIA founder and CEO Jensen Huang said that general-purpose robotics is developing rapidly and the robotics field's "ChatGPT moment" is accelerating its arrival. NVIDIA said this update aims to provide developers with models, data, simulation tools and computing systems to help them apply general-purpose robotics to real industrial scenarios.
The Agile Robots Agile ONE SN uses the Agile ONE S humanoid robot body, standing 1.4 meters tall and weighing 40 kg, with 31 degrees of freedom and a single-arm payload of 3 kg. It is equipped with two WUJI Hand 2 five-finger dexterous hands, each with 20 degrees of freedom, supporting 1kHz force control and tactile sensing. It should be noted that Reuters reported the robot weighs about 34 kg, which differs from NVIDIA's officially stated figure of 40 kg, and available information is not yet sufficient to confirm the reason for the discrepancy.
On the computing side, the Agile ONE SN is equipped with NVIDIA Jetson Thor, delivering 2070 TFLOPS of FP4 computing power and 128GB of unified memory. Sensors include head and wrist cameras, LiDAR and an inertial measurement unit, used to perceive the surrounding environment and support the robot's interaction with the external environment.
On safety, the platform plans to integrate on-robot safety protocols with NVIDIA's Halos full-stack safety system, though this integration is still pending. NVIDIA said that research institutions such as Stanford University and ETH Zurich have expressed interest in the Agile ONE SN.
Reuters also reported that Agile Robots plans to produce up to 2000 humanoid robots in partnership with NVIDIA in 2027, targeting customers such as universities and corporate research labs.
On the software side, NVIDIA Isaac Lab 3.0 has been officially released. As an open-source robot learning framework, it supports multi-GPU training workflows and can use physics engines such as Newton and PhysX. The framework aims to help developers train and optimize robot policies, covering computing environments ranging from desktop workstations to data centers.
IT之家 noted that NVIDIA has also launched the open-source runtime framework Isaac OS. Built on Holoscan, the framework is used to connect AI models, sensors, control systems and robot hardware, supports migrating from simulation environments to physical robots, and provides deterministic real-time execution capabilities.
For the DGX Station, NVIDIA simultaneously released three open-source reference workflows. The first optimizes the GR00T 1.7 model through reinforcement learning; the second uses Isaac Lab to train dexterous manipulation skills; and the third uses ChatGPT Astra to post-train the Cosmos world model to generate synthetic training data. These workflows are designed to help developers integrate robot learning, simulation and data generation stages.
On the data side, the SimReady Central Library offers over 2000 simulation-ready robot and factory assets, covering a variety of robot equipment and industrial scenario resources. The Human 1M dataset contains a variety of human activities, synchronized three-view videos and 3D hand-tracking data, usable for robot learning and motion understanding. Both resources are available through Hugging Face, with the full version of the Human 1M dataset open only to Cosmos Coalition partners.
NVIDIA also introduced several companies adopting its robot development tools. ABB Robotics has integrated the Omniverse library into RobotStudio HyperReality for simulation and virtual commissioning before robot system installation; NEURA Robotics uses Isaac Sim and Isaac Lab to generate synthetic training data and train robot skills in simulated environments. NVIDIA said companies such as KUKA, Humanoid, Sereact and Universal Robots are also adopting technologies related to its open robot platform.