NVIDIA today announced a major expansion of the open-source robotics development platform NVIDIA Isaac ™, adding a new humanoid robot reference prototype, accompanying datasets, and a runtime, while also upgrading the robot learning framework.

Relying on this technology, developers can obtain an integrated solution from data collection to actual deployment, enabling the faster introduction of general-purpose robots into real-world scenarios.

"General-purpose robots are developing at a rapid pace, and the 'golden age' of robotics is just around the corner," said Jensen Huang, founder and CEO of NVIDIA. "With our full-stack open-source robotics platform, we provide developers with models, data, simulation tools, and computing systems to bring physical robotics into the world's major core industries."

Agile ONE SN —— A cutting-edge humanoid robot reference platform for scientific research

Siling Robot ( Agile Robots ) has released Agile ONE SN, which is a full-stack humanoid robot reference prototype pre-installed with NVIDIA Isaac GR00T and equipped with the NVIDIA Jetson Thor ™ Edge AI platform.

Robot researchers and developers often face a major challenge: simulation, training, and deployment tools are disconnected from each other, making it very difficult to transfer robot skills from the development environment to actual robots. Agile ONE SN integrates all these tools and workflows into a single platform, reducing the deployment preparation time from several weeks to just a few days.

Agile ONE SN Parameter Highlights:

• Agile ONE S Humanoid body: height 1.4 meters, weight 40 kilograms, 31 degrees of freedom, maximum load per arm 3 kilograms
• A pair of WUJI Hand five-finger dexterous hands: each hand has 20 degrees of freedom, 1kHz force control, and tactile perception, enabling precise operations.
• Onboard NVIDIA Jetson Thor computing power: FP4 with 2070 TFLOPS of computing power, and a unified memory of 128GB, used for onboard reasoning and real-time control.
• Multimodal perception: Head and wrist cameras, lidar, inertial measurement units to perceive and interact with the environment.
• NVIDIA Isaac GR00T Platform: Includes GR00T large-scale basic models, a complete workflow from data collection to deployment, and full-body coordinated motion control.
• Security capabilities: Plans to integrate with the robot-side security protocol, connecting to NVIDIA's full-stack security system for robots Halos

The breakthrough for humanoid robots lies in their scalability and practicality: a machine that can adapt to various tasks and work in different environments designed for humans. Chen Zhaopeng, the founder of Singularity Robotics and also CEO, stated, "Today, Singularity is launching Agile ONE SN, a full-stack open-source humanoid reference prototype based on NVIDIA Isaac GR00T and equipped with Jetson Thor. We provide developers with a ready-made system to develop, train, and verify the capabilities required for humanoid robots to be put into practical use."

Agile ONE SN is expected to go public in early 2027, focusing on the development of cutting-edge humanoid robots. Top institutions such as Stanford University and the Swiss Federal Institute of Technology Zurich have expressed their interest in collaborating.

NVIDIA’s cutting-edge technology accelerates robot learning: from data to practical application

NVIDIA's new set of toolchains helps developers generate training data, scale up robot learning, build simulation environments, and deploy robot applications, thereby shortening the R&D cycle.

• SimReady Central Resource Library: Provides over 2,000 robots and industrial assets that can be directly used for simulation; Human A 1M dataset contains a rich variety of human actions, synchronized multi-view videos, and 3D hand tracking data. Both have been made available on Hugging Face; The complete Human 1M dataset is only accessible to Cosmos alliance partners.
• NVIDIA Isaac Lab 3.0 officially launched: An open-source agent learning framework that supports parallel workflows from local workstations to data centers, utilizing Newton and NVIDIA PhysX ™ physical engine for robot strategy training and iteration.
• NVIDIA Isaac OS: An open-source runtime framework built on NVIDIA Holoscan, which connects AI, sensors, control units with robot hardware, to achieve deterministic real-time scheduling from simulation to physical robots.
• NVIDIA releases three sets of open-source reference workflows for DGX Station ™: based on reinforcement learning to iterate the GR00T 1.7 model, train dexterous operation skills in Isaac Lab, fine-tune the NVIDIA Cosmos ™ world model, and combine it with ChatGPT Astra to generate synthetic data.
• NVIDIA Halos: A full-stack security platform based on NVIDIA IGX Thor ™, helping autonomous vehicles and robots to move from prototypes to safe, large-scale implementation.

Hugging Face's SimReady repository assets come from: Comau (Comau), Fanuc (FANUC), ForgeCAD, Franka Robotics, LG Electronics, Palatial, Robotiq, Standard Bots, YOLO Robotics (Universal Robots), Vertiv.

NVIDIA's open-source robot learning reference workflow, which runs on DGX Station, is supported by manufacturers such as 1X, Singularity Robotics, Humanoid, Maven Robotics, Mentee, MicroAGI, Rhoda, Cyberkin ( Sereact ), and Skild AI.

Multiple robotics partners have integrated Halos into their security architectures: Singularity Robotics, Agility, Dexory, FORT Robotics, KION for warehouse robot security solutions; SICK ( SICK ) is evaluating Halos to develop dynamic security capabilities for industrial robots, breaking through the limitations of fixed security zones to address dynamically changing scenarios.

Advantech and NexCOBOT are developing industrial computing platforms based on NVIDIA IGX; Reynolds & Moore, Accenture, and TÜV SÜD are accelerating the certification process for industrial physical AI security deployments.

European robotics manufacturers rely on NVIDIA's open-source technology for research and development.

ABB The robot will integrate the NVIDIA Omniverse™ library with RobotStudio HyperReality, and complete robot system simulation and virtual debugging before production, covering multiple industries. KUKA Group (KUKA Group) integrates Omniverse components within Visual Components FactoryLens, adding real-time rendering to factory simulations. Cerec (Sereact) relies on Isaac GR00T + Cosmos Transfer to train and fine-tune a Cortex platform for cargo sorting, which is then deployed on Yuao industrial robotic arms.

Relying on Jetson Thor, Holoscan Sensor Bridge, Halos, and Isaac OS, robot behavior development and verification are completed before industrial trials; Hexagon Robotics focuses on developing the perception capabilities of AEON, which include teleoperation video, positioning, mapping, and object pose detection; Bear Robotics accelerates the perception and model execution of KR1, with the assistance of Kino-2 for fine-tuning of Isaac GR00T, targeting warehousing, return goods, and industrial scenarios.

Toyota Material Handling Europe, SoftServe, uses Omniverse + NVIDIA Isaac Sim ™ to develop its own material handling solutions, reducing deployment times by up to 50%. Idealworks relies on NVIDIA's open-source models and Jetson to equip factory autonomous mobile robots with memory capabilities. ADI > ADNOC ( ADI ) utilizes Jetson AGX Orin ™ and the cuVSLAM library to develop its own autonomous mobile robots.

Hugging Face and Pollen Robotics are built upon Isaac Lab, Isaac Sim, and Newton to create Microduck, a small open-source robot for developers and students to train, modify, and share new embodied behaviors. Franka Robotics integrates with NVIDIA NIM™ microservices + Isaac Sim to convert natural language commands into Franka FR3 Duo robotic arm movements. Relying on Jetson and GR00T, Billie drives the robot, which serves as its first physical AI robotic partner and has been deployed in hotels under Marriott, Hilton, InterContinental, and OYO.

Flexion uses Isaac Lab and Newton for the scaled training of the unified robot intelligence model Reflect; Reflect integrates reasoning with full-body movement to perform complex operations across different models. Schaeffler (Schaeffler) provides exclusive actuator technology to Newton, in conjunction with Isaac Lab and Isaac Sim, to accelerate the development of humanoid robots.

NEURA Robotics uses Isaac Sim and Isaac Lab to generate synthetic training data, training robot skills for the multimodal base model Neuron AI in a simulation environment.

Responsible Editor: Liu Mingliang