NVIDIA Showcases Physical AI Breakthroughs and New Robot Models at National Robotics Week

NVIDIA used National Robotics Week 2026 to showcase a sweeping set of advances in physical AI, releasing new open models, simulation tools, and an edge-to-cloud orchestration framework designed to accelerate the path from robotic research to real-world deployment.

New Open Models for Robot Learning

At the center of the announcement were updates to Cosmos and GR00T, NVIDIA's open foundation models for physical AI. Cosmos generates photorealistic synthetic environments that robots can use for training, while GR00T provides generalized locomotion and manipulation capabilities for humanoid robots. Both models received significant architecture upgrades, with NVIDIA reporting improvements in sample efficiency and sim-to-real transfer performance.

The open release strategy is deliberate. By making these models freely available to researchers and developers, NVIDIA is positioning its hardware and simulation platforms as the default infrastructure for the emerging robotics industry, mirroring the playbook it used to dominate AI training.

Simulation and Evaluation Tools

NVIDIA introduced Isaac Lab-Arena, a standardized evaluation framework built on its Isaac Lab simulation environment and Omniverse platform. Isaac Lab-Arena provides reproducible benchmarks for comparing robot learning algorithms, addressing a longstanding gap in the field where inconsistent evaluation methods have made it difficult to measure genuine progress.

The company also launched RoboLab, a simulation benchmark suite that tests robots across manipulation, navigation, and multi-agent coordination tasks. Both tools run on NVIDIA GPUs and integrate with the broader Isaac and Omniverse ecosystems.

Complementing these simulation tools, NVIDIA unveiled OSMO, an edge-to-cloud orchestration framework that manages the lifecycle of robotic AI models from training in the cloud to deployment on edge devices in factories, warehouses, and other physical environments.

Industry Partners Debut New Robots

Several of NVIDIA's robotics partners used the event to debut new hardware. Boston Dynamics demonstrated updated Atlas capabilities powered by NVIDIA compute. Caterpillar showed autonomous construction equipment running NVIDIA's Jetson platform. Franka Robotics, LG Electronics, and NEURA Robotics each presented next-generation systems leveraging the new Cosmos and GR00T models for tasks ranging from precision manufacturing to household assistance.

Accelerating Real-World Deployment

The combined announcements reflect NVIDIA's bet that the convergence of simulation, synthetic data, and foundation models will compress the timeline for deploying capable robots from decades to years. CEO Jensen Huang has repeatedly described physical AI as the next trillion-dollar computing platform, and the scale of investment NVIDIA is directing at the space suggests the company is positioning itself as the essential infrastructure provider for that transition.