NVIDIA

Overview

NVIDIA goes beyond GPUs to provide core infrastructure for robotics AI. They are building a full-stack robotics solution spanning simulation (Omniverse), world models (Cosmos), humanoid models (GR00T), development platforms (Isaac), and edge computing (Jetson).

Item	Details
Headquarters	Santa Clara, CA
CEO	Jensen Huang
Robotics Research Lab	GEAR Lab (Generalist Embodied Agent Research)
Core Products	Isaac, Omniverse, Cosmos, GR00T, Jetson

GEAR Lab

The core organization leading NVIDIA’s Physical AI research.

Item	Details
Full Name	Generalist Embodied Agent Research
Co-Leaders	Jim Fan, Yuke Zhu
Research Areas	Foundation Models, World Models, Embodied AI

Key People

Jim Fan: NVIDIA Robotics Director & Distinguished Scientist. Stanford Ph.D., OpenAI’s first intern
Yuke Zhu: Associate Professor at UT Austin, NVIDIA Director & Distinguished Research Scientist. Leads RPL Lab

Isaac Platform

An end-to-end platform for robot development.

Isaac Sim

Feature	Description
Physics Engine	PhysX 5 based
Rendering	RTX ray tracing
Sensor Simulation	Camera, LiDAR, IMU
Integration	ROS 2, USD support

Isaac Lab

Reinforcement learning/imitation learning training environment:

GPU-parallel simulation
Domain Randomization
Sim-to-Real pipeline

Isaac Lab-Arena (2026)

Simulation evaluation framework announced at CES 2026:

Integrates Libero, RoboCasa, RoboTwin benchmarks
Virtual validation of robot capabilities

Omniverse

A physically accurate digital twin platform.

Core Features

Feature	Description
USD	Universal Scene Description
RTX	Real-time ray tracing
PhysX	Physics simulation
Nucleus	Collaborative data layer

Robotics Applications

[Real Factory] ──3D Scan──→ [Digital Twin] ──Simulation──→ [Robot Training]
                                  ↓
                            [Real Deployment]

Cosmos

World Foundation Model

World model platform announced at CES in January 2025.

Item	Details
Purpose	World Model for Physical AI
Training Data	20M hours of video (9,000 trillion tokens)
Model Types	Nano, Super, Ultra
Applications	Robotics, autonomous driving

Core Capabilities

Physical Future Prediction: Predicting what happens next
Synthetic Data Generation: Training data augmentation
Reasoning: Simulating action consequences

Jensen Huang: “The ChatGPT moment for robotics is coming”

GR00T (Project)

Humanoid Foundation Model

Item	Details
Announcement	GTC 2024 (March 2024)
Name Meaning	Generalist Robot 00 Technology
Target	Humanoid robots
Architecture	Dual-System (System 1 + System 2)

Dual-System Architecture

Inspired by principles of human cognition:

System 1: Fast-thinking (actions/skills) - Diffusion Transformer
System 2: Slow-thinking (reasoning/planning) - Vision-Language Model

GR00T N1 (March 2025)

Open-source humanoid VLA announced at GTC 2025:

Vision-Language-Action model
Cross-embodiment support
Available on Hugging Face (nvidia/GR00T-N1-2B)

GR00T N1.5 (June 2025)

Announced at COMPUTEX 2025:

GR00T-Dreams synthetic data generation
RTX PRO 6000 system integration

GR00T N1.6 (September 2025)

Announced at CoRL 2025:

Cosmos Reason integration
Enhanced reasoning capabilities

Newton Physics Engine

Open-Source Physics Engine

Item	Details
Announcement	GTC 2025 (March 2025)
Co-Developed By	NVIDIA, Google DeepMind, Disney Research
Governance	Linux Foundation
Built On	NVIDIA Warp, OpenUSD

Performance

152x faster than MuJoCo MJX for humanoid locomotion
313x faster for manipulation tasks (on RTX 4090)

Jetson Hardware

Jetson Thor

Item	Details
Announcement	GTC 2024 (March 2024)
Availability	August 2025
Architecture	Blackwell GPU
Performance	2070 TFLOPS (FP4)
Memory	128GB
Power	40W - 130W

Pricing

Developer Kit: $3,499
T5000 module: $2,999 (at 1,000 units)

Jetson T4000 (2026)

Announced at CES 2026:

Blackwell architecture based
$1,999 (at 1,000 units)
Upgrade path for Jetson Orin customers

Partnerships

Humanoid Partners

Company	Robot	Collaboration
Boston Dynamics	Atlas	Jetson Thor integration announced
Figure AI	Figure 02	Using Isaac/Omniverse
Agility Robotics	Digit	Jetson Thor planned for Digit Gen 6
1X Technologies	NEO	GR00T platform
Apptronik	Apollo	Isaac simulation
Unitree Robotics	-	GR00T partner
XPENG Robotics	-	GR00T partner
Sanctuary AI	-	GR00T N1 early access
Fourier Intelligence	-	Cosmos early adopter

Automotive/Industrial Partners

Mercedes-Benz (factory digital twin)
BMW (manufacturing simulation)
Johnson & Johnson MedTech (surgical robot simulation)

Research Partners

Google DeepMind (Newton co-development)
Disney Research (Newton co-development, robot characters)
Hugging Face (LeRobot integration)

Strategic Position

                  ┌─────────────────┐
                  │     Cosmos      │ ← World Model
                  │ (Physical Pred) │
                  └────────┬────────┘
                           │
┌──────────┐      ┌────────▼────────┐      ┌──────────┐
│  Isaac   │──────│   Omniverse     │──────│  GR00T   │
│  (SDK)   │      │  (Simulation)   │      │ (Model)  │
└──────────┘      └────────┬────────┘      └──────────┘
                           │
                  ┌────────▼────────┐
                  │  Jetson Thor    │ ← Hardware
                  │ (Edge Inference)│
                  └─────────────────┘

NVIDIA’s strategy is to dominate the infrastructure layer of robotics AI. They have 2M+ robotics developers and 150+ partner ecosystem.