NVIDIA today announced a major expansion of the open-source robotics development platform NVIDIA Isaac™, introducing a new humanoid robot reference design, accompanying datasets and runtime, along with an upgraded robotics learning framework.
With this technology, developers can access an integrated solution from data collection to deployment, accelerating the real-world application of general-purpose robots.
"General-purpose robots are advancing rapidly, and the 'ChatGPT moment' for robotics is just around the corner," said Jensen Huang, founder and CEO of NVIDIA. "With the fully open-source robotics platform NVIDIA Isaac, we are providing developers with models, data, simulation tools, and computing systems to bring physical AI into major industries worldwide."
Agile ONE SN — A cutting-edge humanoid robot reference platform for research
Agile Robots has launched the Agile ONE SN, a full-stack humanoid robot reference platform preloaded with NVIDIA Isaac GR00T and powered by the NVIDIA Jetson Thor™ edge AI platform.
Robot developers often face a major pain point: simulation, training, and deployment tools are fragmented, making it difficult to transfer robotic skills from the development environment to physical robots. Agile ONE SN integrates all tools and workflows into a single platform, reducing deployment preparation time from weeks to just days.
Agile ONE SN Key Features:
- Agile ONE S Humanoid Body: Height 1.4 meters, weight 40 kg, 31 degrees of freedom, maximum payload of 3 kg per arm
- A pair of WUJI Hand 2 five-finger dexterous hands: each hand features 20 degrees of freedom, 1 kHz force control, and tactile sensing for precise manipulation.
- Onboard NVIDIA Jetson Thor computing power: 2070 TFLOPS FP4, 128GB unified memory for onboard inference and real-time control.
- Multimodal perception: Head and wrist cameras, LiDAR, and inertial measurement units sense and interact with the environment.
- NVIDIA Isaac GR00T Platform: Includes the GR00T foundational large model, a complete workflow from data collection to deployment, and whole-body coordinated motion control.
- Security Capabilities: Plan to integrate with the bot-side security protocol and connect to NVIDIA’s full-stack security system, Halos, designed for bots.
“The breakthrough for humanoid robots lies in scalable practicality: an adaptive machine that learns multiple skills and works in environments designed for humans,” said Chen Zhaopeng, Founder and CEO of Siasun Robotics. “Today, Siasun introduces the Agile ONE SN—a fully open-source humanoid reference platform based on NVIDIA Isaac GR00T and powered by Jetson Thor—providing developers with a ready-made system to develop, train, and validate the capabilities needed for real-world humanoid robot deployment.”
Agile ONE SN is expected to launch in early 2027, targeting advanced humanoid robotics research, with leading institutions such as Stanford University and ETH Zurich expressing interest in collaboration.
NVIDIA's new technology accelerates robot learning: from data to deployment
NVIDIA's new toolchain helps developers generate training data, scale robotic learning, build simulation environments, and deploy robotic applications, shortening the development cycle.
- SimReady Central Repository: Offers over 2,000 robots and industrial assets ready for direct simulation; the Human 1M dataset includes rich human motion, synchronized multi-view video, and 3D hand tracking data. Both are now available on Hugging Face; the full Human 1M dataset is accessible exclusively to Cosmos Alliance partners.
- NVIDIA Isaac Lab 3.0 is now officially available: an open-source agent learning framework supporting workflows from local workstations to data centers with multi-GPU parallelism, enabling training and iteration of robotic policies using Newton and NVIDIA PhysX™ physics engines.
- NVIDIA Isaac OS: An open-source runtime framework built on NVIDIA Holoscan that integrates AI, sensors, control units, and robotic hardware to enable deterministic real-time scheduling from simulation to physical robots.
- NVIDIA has released three open-source reference workflows for DGX Station™: iteratively refining the GR00T 1.7 model using reinforcement learning, training dexterous manipulation skills in Isaac Lab, and fine-tuning the NVIDIA Cosmos™ world model combined with ChatGPT Astra to generate synthetic data.
- NVIDIA Halos: A full-stack secure platform based on NVIDIA IGX Thor™, enabling autonomous vehicles and robots to move from prototype to safe, scalable deployment.
The SimReady resources on Hugging Face are sourced from: Comau, FANUC, ForgeCAD, Franka Robotics, LG Electronics, Palatial, Robotiq, Standard Bots, Universal Robots, and Vertiv.
The NVIDIA open-source robotics learning reference workflow, running on DGX Station, is supported by companies including 1X, Siasun Robotics, Humanoid, Maven Robotics, Mentee, MicroAGI, Rhoda, Sereact, and Skild AI.
Multiple robotic partners have integrated Halos into their safety architectures: Siasun, Agility, Dexory, FORT Robotics, and KION, for warehouse robot safety solutions; SICK is evaluating Halos to enable dynamic safety capabilities for industrial robots, moving beyond fixed safety zones to adapt to changing environments.
Advantech and NexCOBOT are developing an industrial computing platform based on NVIDIA IGX; Reynolds & Moore, Accenture, and TÜV SÜD are accelerating the certification process for secure deployment of industrial physical AI.
European robotics manufacturers are leveraging NVIDIA's open-source technologies for research and development.
ABB robots integrate the NVIDIA Omniverse™ library into RobotStudio HyperReality to simulate and virtually commission robotic systems before deployment, covering multiple industries. KUKA Group integrates Omniverse components into Visual Components FactoryLens to add real-time rendering to factory simulations. Sereact trains and fine-tunes its Cortex platform for cargo sorting using Isaac GR00T + Cosmos Transfer, deploying it on Universal Robots industrial arms.
Humanoid leverages Jetson Thor, Holoscan Sensor Bridge, Halos, and Isaac OS to develop and validate robotic behaviors prior to industrial pilot deployment; Hexagon Robotics develops AEON perception capabilities, including teleoperated video, localization, mapping, and object pose detection; Bear Robotics accelerates perception and model execution for KR1 by fine-tuning Isaac GR00T with Kino-2 for warehouse, return, and industrial use cases.
Toyota Material Handling Europe and SoftServe use Omniverse and NVIDIA Isaac Sim™ to develop autonomous material handling solutions, reducing deployment time by up to 50%. Idealworks leverages NVIDIA’s open-source models and Jetson to enable memory capabilities for factory autonomous mobile robots. Analog Devices (ADI) uses Jetson AGX Orin™ and the cuVSLAM library to develop autonomous mobile robots.
Hugging Face and Pollen Robotics have developed Microduck, a small open-source robot built on Isaac Lab, Isaac Sim, and Newton, enabling developers and students to train, modify, and share novel embodied behaviors. Franka Robotics integrates NVIDIA NIM™ microservices with Isaac Sim to convert natural language instructions into actions for the Franka FR3 Duo robotic arm. Bellboy leverages Jetson and GR00T to power Billie, its first physical AI robotic companion, now deployed in hotels under Marriott, Hilton, InterContinental, and OYO.
Flexion scales training for the unified robotic intelligence model Reflect using Isaac Lab and Newton; Reflect integrates reasoning with whole-body motion to perform complex tasks across different robot platforms. Schaeffler exclusively provides actuator technology to Newton, accelerating humanoid robot development in conjunction with Isaac Lab and Isaac Sim.
NEURA Robotics uses Isaac Sim and Isaac Lab to generate synthetic training data and train robotic skills for the multimodal foundation model Neuron AI in a simulated environment.
Editor-in-Chief: Liu Mingliang