Learning path
NVIDIA Cosmos : Foundations of Physical AI
Skill Level
Introductory
Duration 1 hour 20 minutes
Updated Sep 30, 2026
About this learning path
Robots and autonomous vehicles need AI that understands the physical world, not just text. NVIDIA Cosmos is a family of world foundation models built for exactly that. This learning path gives you the vocabulary and the mental model: what physical AI is, how Cosmos works, where it fits in a real workflow, and what it takes to run it. No robotics background needed.
Your instructors
Prerequisites
- General AI and ML terms: what a foundation model is, and what training data and inference mean
- Basic familiarity with containers and GPUs (helpful, not required)
What you'll learn
- Define physical AI and how it differs from generative AI that works only with language
- Identify the three Cosmos model families: Predict, Transfer, and Reason
- Explain how Cosmos represents and generates video through tokenization
- Recognize core use cases: synthetic data, sim-to-real transfer, and policy training
- Name the building blocks needed to self-host Cosmos: GPUs, memory, a container runtime,NIM, and Kubernetes
- Find the resources to keep learning: NGC, NIM, GitHub, and the Cosmos Coalition