NVIDIA expands Isaac tools with Agile ONE SN humanoid platform
NVIDIA has expanded its Isaac robotics development platform with training tools, datasets and a runtime framework, alongside Agile Robots’ announcement of the Agile ONE SN humanoid reference platform. According to engineering.com, the robot combines NVIDIA Isaac GR00T with Jetson Thor onboard computing and is scheduled to become available in early 2027.
The package addresses a practical engineering task in humanoid development: connecting data capture, simulation, policy training and physical robot operation. Agile ONE SN packages those functions around a specified humanoid body and dexterous hands, rather than leaving developers to assemble the entire hardware and software stack themselves.
A 40 kg humanoid for development work
The platform uses the Agile ONE S body, which stands 1.40 meters tall, weighs 40 kilograms and has 31 degrees of freedom. Its rated payload is 3 kilograms per arm. Two WUJI Hand 2 hands each provide five fingers, 20 degrees of freedom, tactile sensing and 1 kHz force control.
Head and wrist cameras, lidar sensors and inertial measurement units supply perception data. Jetson Thor provides 128 GB of unified memory and a stated 2,070 FP4 teraflops of compute. The GR00T integration includes a foundation model, a workflow covering data capture through deployment, and whole body control for coordinated movement.
Agile Robots plans to integrate NVIDIA Halos safety capabilities into the platform. Stanford University and ETH Zurich are among the institutions expressing interest in the research and development system, engineering.com reported.
Training and runtime tools join the hardware
NVIDIA says Isaac Lab 3.0 is generally available as an open source framework for training and refining robot policies. It supports multiple GPUs and physics engines including Newton and NVIDIA PhysX. The new Isaac OS, built on Holoscan, connects AI systems, sensors, controls and robot hardware for real time operation in simulation and on physical robots.
The data additions include the Human 1M Dataset, with human activities, synchronized video from three views and 3D hand tracking. NVIDIA is making it available on Hugging Face, while reserving the complete dataset for Cosmos Coalition partners. A separate SimReady Central Library contains more than 2,000 simulation assets for robotics and factories.
NVIDIA is also releasing reference workflows for refining GR00T 1.7 models with reinforcement learning and training dexterous manipulation skills in Isaac Lab. These target skill development rather than introducing another humanoid body.
Among humanoid developers using the stack, Humanoid is employing Jetson Thor, Holoscan Sensor Bridge, Halos and Isaac OS to develop and validate behaviors before industrial pilot projects. Hexagon Robotics is using the same technologies for AEON perception capabilities, including teleoperation video, localization, mapping and object pose detection.
Source: Engineering.com
