Key Takeaways

• NVIDIA expands the Isaac open‑source robotics stack with a full‑stack humanoid reference platform, new datasets, and a runtime framework.
• Agile ONE SN combines NVIDIA Isaac GR00T, Jetson Thor AI compute, and dual WUJI Hand 2 manipulators into a single development kit.
• The platform delivers 31 DoF, 1.4 m height, 40 kg mass, and 2,070 FP4 TFLOPs of edge compute, targeting research labs and early‑stage product teams.
• New data assets – SimReady Central Library (2,000+ assets) and Human 1M (1 M synchronized human activity clips) – accelerate simulation‑to‑real pipelines.
• Availability is slated for early 2027, with early adopters including Stanford University and ETH Zurich.

NVIDIA Broadens Isaac for Humanoid Robots

NVIDIA announced a major upgrade to its Isaac open‑source robotics platform, adding a humanoid reference system, enriched data libraries, and a unified runtime for robot learning. The end‑to‑end workflow now spans data generation → high‑fidelity simulation → AI training → real‑world deployment, reducing the engineering overhead that traditionally hampers humanoid development.

Agile ONE SN: A Turnkey Humanoid Reference Platform

Agile Robots unveiled Agile ONE SN, a full‑stack humanoid built around the NVIDIA Isaac GR00T foundation model and powered by the Jetson Thor edge AI module. The system bundles simulation, training, and deployment tools on a single hardware package, allowing researchers to focus on skill development rather than integration.

Core Specifications

The table highlights Agile ONE SN’s higher DOF, lighter weight, and superior on‑board compute, positioning it as a research‑focused alternative to larger commercial humanoids.

Key Hardware Elements

• Agile ONE S Body – 1.4 m tall, 40 kg, 31 DoF, and a 3 kg payload per arm, enabling complex manipulation and locomotion tasks.
• Dual WUJI Hand 2 – Five‑finger hands delivering 20 DoF, 1 kHz force control, and tactile feedback for dexterous grasping.
• Jetson Thor – Delivers 2,070 FP4 TFLOPs and 128 GB unified memory, supporting real‑time perception, planning, and whole‑body control.
• Multimodal Sensing – Combines head‑mounted RGB cameras, wrist‑mounted RGB cameras, lidar, and IMU for robust environment understanding.

Expanded Isaac Toolset for Robot Learning

NVIDIA introduced several data‑centric and simulation assets to accelerate robot‑learning pipelines:

• SimReady Central Library – Over 2,000 simulation‑ready assets (robots, fixtures, factory equipment) optimized for Isaac Lab and Omniverse.
• Human 1M Dataset – A curated collection of 1 million human activity clips, each with synchronized three‑view video and 3‑D hand tracking. The dataset is hosted on Hugging Face and fully accessible to members of the Cosmos Coalition.
• Isaac Lab 3.0 – The latest open‑source release, featuring GPU‑accelerated physics, reinforcement‑learning environments, and a unified API for rapid prototyping.

These resources enable developers to generate realistic training data, scale learning workloads on NVIDIA GPUs, and transition policies from simulation to physical robots with minimal domain‑gap.

Timeline & Early Adoption

• Q1 2027 – First shipments of Agile ONE SN to research partners.
• Pilot Institutions – Stanford University and ETH Zurich have signed memoranda of understanding to evaluate the platform for bipedal locomotion and dexterous manipulation studies.

Bottom Line

NVIDIA’s expansion of the Isaac ecosystem, anchored by the Agile ONE SN humanoid reference platform, delivers a cohesive hardware‑software stack that dramatically lowers the barrier to humanoid robot research. With industry‑leading compute, high‑DOF manipulators, and a rich suite of simulation and data assets, the solution is poised to become the go‑to platform for universities and early‑stage innovators targeting real‑world robot learning. The early‑2027 launch gives the research community a timely, future‑proof toolset to accelerate the transition from simulation to safe, capable humanoid robots.