NVIDIA Isaac & Physical AI Platform

NVIDIA Isaac is NVIDIA's end-to-end platform for developing, training, and deploying physical AI — the software and model stack that an increasing majority of the humanoid robot industry builds on. Isaac encompasses simulation (Isaac Sim), foundation models (GR00T for humanoid control, Cosmos for world models), training frameworks (Isaac Lab), and edge deployment (Jetson). NVIDIA's ambition, as described by CEO Jensen Huang, is to make Isaac "the Android of robotics" — the standard platform that every robot developer uses, just as Android became the standard for smartphones.

GR00T: Humanoid Foundation Models

GR00T (Generalist Robot 00 Technology) is the world's first open, commercially licensable foundation model for humanoid robot control. GR00T N1 features a dual-system architecture: System 2 is a vision-language model that reasons about the environment and plans actions ("pick up the box, walk to the shelf, place it"); System 1 is a fast action model that translates plans into continuous motor commands at hundreds of hertz. The dual system mirrors human cognition: deliberate reasoning for planning, reflexive control for execution.

GR00T N1.7 (March 2026) shipped with advanced dexterous control and commercial licensing, making production-ready humanoid AI available to any robot manufacturer. GR00T N2, previewed at GTC 2026, is based on DreamZero research and integrates world model capabilities — the robot can imagine the consequences of actions before executing them.

Cosmos: World Foundation Models

Cosmos world foundation models generate photorealistic predictions of future visual states, serving dual roles. As data generators, Cosmos Transfer takes a small number of real human demonstrations and produces exponentially more synthetic variations — different objects, lighting, viewpoints — that maintain physical plausibility. NVIDIA demonstrated generating 780,000 synthetic trajectories (equivalent to 9 months of human demonstrations) in 11 hours. As planning modules, Cosmos models enable robots to predict what will happen if they take a proposed action, enabling safer and more deliberate physical interaction.

Isaac Sim and Isaac Lab

Isaac Sim, built on NVIDIA Omniverse, provides the physics simulation environment for sim-to-real transfer. It simulates robot dynamics, sensor models, contact physics, and visual rendering with GPU acceleration, enabling thousands of parallel training environments to run simultaneously. Isaac Lab provides the reinforcement learning and imitation learning frameworks for training robot policies within Isaac Sim. Together, they form the complete training pipeline: simulate the world, train the policy, deploy to hardware.

The Platform Strategy

NVIDIA's robotics strategy mirrors its AI strategy: don't build the robots, build the platform everyone needs to build robots. Every major humanoid company — Figure AI, Agility Robotics, Apptronik, Unitree, Boston Dynamics — uses some combination of Isaac Sim, GR00T, Cosmos, and NVIDIA Jetson for edge inference. This positions NVIDIA to capture value from the entire humanoid robotics wave regardless of which hardware company wins, just as NVIDIA captured value from the AI training wave regardless of which model company led. The robotics platform is also a demand driver for NVIDIA's core business: every simulated training environment runs on NVIDIA GPUs, and every deployed robot runs inference on NVIDIA edge hardware.