RoboBoom
Nebius· December 26, 2025· 12:31

Amit Goel (NVIDIA) – The Future of Robotics & Physical AI | Nebius Robotics Summit 2025

Amit Goel of NVIDIA explains the opportunities, technical challenges, and development stack for physical AI at the Nebius Robotics Summit 2025.

In this talk

  1. 0:14

    Physical AI expands automation into the real world

    Goel describes a progression from perception and generative AI to agentic AI, then identifies physical AI as the next wave. He sees a large opportunity in factories, hospitals, stores, warehouses, and transportation, especially as manufacturing and onshoring become priorities. He estimates the existing physical-world opportunity at tens of trillions of dollars and says AI could expand it further.

  2. 2:52

    Physical AI needs data, simulation, and edge computing

    Goel says physical AI lacks the abundant training data available to language models, and real-world testing brings cost and safety risks. Systems also need to interpret 3D space, time, force, audio, and touch, alongside images and text. Low-latency edge computing is necessary to deploy models across physical devices.

  3. 5:06

    A three-computer platform supports the development cycle

    Goel outlines a simulation computer using Omniverse, Cosmos, and RTX for generating data and testing models; a Blackwell AI factory for training; and edge systems for deployment. He says NVIDIA is building software, models, libraries, and frameworks across these layers, and that much of the stack can run on Nebius.

  4. 8:19

    Isaac tools help train and validate robots

    Isaac Lab supports robot learning across tens of thousands of simulated environments, varying properties such as mass and friction so robots can gain experience in parallel on GPU clusters. Goel says many quadrupeds and bipeds learn locomotion in Isaac Lab or Isaac Gym. Isaac Sim lets developers bring in robots and environments to test and validate systems before real-world deployment.

  5. 10:08

    Startups apply simulation to dexterity and industry

    Goel names General Robotics, which offers reusable environments, assets, and models for testing, and Korea-based Real World, which uses GR00T and Isaac Lab to tackle dexterity with high-degree-of-freedom arms and hands. Worker Labs applies Isaac Lab and Isaac Sim in industrial automation, including pick-and-place, CNC, and precision tasks. He also invites founders to join NVIDIA's physical AI ecosystem and startup programs.

Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.

More talks