Continuing with my new protocol, here I’m going to share content I consumed and learned from. This week, I spent time trying to understand physical AI, whether it’s the next frontier in AI, and how valuable training data sets are in training models for physical AI.

What I consumed this week and what I learned from it:

• Closing the data gap in robotics – YouTube presentation by UC Berkeley Ken Goldberg at an MIT Robotics seminar. This helped me understand the role that data plays in training models for robotics. His framing and quantification of the robot data-gap problem in hours and then 100,000 years was eye opening. His categorization of the four current ways to solve this problem was helpful too.
• NVIDIA Cosmos world foundation models for physical AI – YouTube presentation by Ming-Yu Liu at NVIDIA’s March 2026 GTC conference explaining their model for generating the photorealistic data needed to train machines and robots to navigate the real world. This helped me understand what Cosmos is and what it does.
• Introducing NVIDIA’s Cosmos world models – YouTube presentation by NVIDIA’s VP Generative AI Research, Ming-Yu Liu at NVIDIA’s March 2025 GTC conference. This is their introduction of Cosmos world models. This helped me understand how Cosmos was introduced last year and how far it’s come in a year.
• NVIDIA’s robotics research – YouTube presentation by Jim Fan, Yashraj Narang, and Yuke Zhu about NVIDIA’s latest robotics research. This video helped me understand NVIDIA’s robotics mission and how its strategy for physical AI focuses on offering three “computers” to customers to help them solve physical AI. Understanding their lab’s three pillars (AI, simulation, and compute-focused ) for solving physical AI research problems was helpful.
• Digital twins and simulations – YouTube presentation by Rev Lebaredian at NVIDIA’s 2026 GTC conference about how digital twins and simulation will help make physical AI a reality. This helped me understand why physical AI is so important to NVIDIA, what each of the three computers needed for robotics actually does, and how digital twins and simulations fit into that model.
• Intro to physical AI and robotics – YouTube presentation by Kalyan Vadrevu on what physical AI and robotics are. This is a good introduction. It helped me understand what physical AI really is and why NVIDIA things it’s the next wave of AI.

That’s what I consumed and learned from last week.