Robots and 'World Models'

Описание: We're starting to explore world models. And we found it critical because as you start evaluating models on hundreds, thousands of different scenarios, it becomes really time consuming to even run the eval, let alone collect data for it.
--
Fei-Fei Li's new company 'World Labs' raises $1B in funding to develop world foundation models.

The investment community has finally figured it out.
Robots are AI unbound from the confines of the computer or mobile phone.
And are out in the world, doing things, interacting with the environment.

World models are an avenue to get more training, faster.
Easier to run a kajillion trials in a digital world, than a physical one.

Foundation models are changing robotics.

AI pioneer Fei-Fei Li's World Labs raises $1 billion in funding.
https://www.reuters.com/business/ai-p...

This interview is co-released by Turn the Lens and Humanoids Summit. Humanoids Summit is organized and hosted by ALM Ventures. Recorded at the Humanoids Summit SV 2025, Computer History Museum, Mountain View, California.

For more on Humanoids Summit, including May 2026 Summit in Tokyo visit
https://humanoidssummit.com/

For more from Humanoids Summit SV
   • 🎙️ From Computer Vision to Robotics | Inte...  

Carolina Parada: Embodied AI, Gemini Robotics, Delightful Surprise | Turn the Lens with Jeff Frick Ep44
Transcript, show notes, and extensive links and resources
https://www.turnthelenspodcast.com/ep...
YouTube -    • Carolina Parada: Embodied AI, Gemini Robot...  
Spotify - https://open.spotify.com/episode/66Np...
or where ever you podcast.

In this clip from Turn the Lens, Carolina Parada (Senior Director at Google DeepMind) explains the shift from manual data collection to autonomous simulation. If we want robots to work in the real world, we have to change how they learn.

The current state of robotics relies heavily on teleoperator data—humans manually showing robots how to translate understanding into action. While this works for teaching a model to operate a new robot, it doesn't scale.

Why are World Models the "Game Changer"?

The Scale Problem: Evaluating a robot on hundreds or thousands of different scenarios manually is too time-consuming.

Speed: World models allow AI to run "evals" in simulation at speeds impossible in the physical world.

Safety: It is critical to test dangerous or unpredictable scenarios in a world model before putting a robot in a real human environment.

Carolina discusses why exploring diverse data sources is the key to moving past today’s limitations. By leveraging these world models, we can evaluate thousands of scenarios and ensure safety without the massive time sink of traditional data collection.

This is a deep dive into the technical infrastructure making "Embodied AI" possible.
🔗 WATCH THE FULL INTERVIEW HERE:    • Carolina Parada: Embodied AI, Gemini Robot...  

Robotics data bottleneck, Google DeepMind research, Carolina Parada, World Models in AI, Teleoperation vs Simulation, Scaling robotics data, AI Safety and Evaluation, Embodied AI development, Machine Learning for Robotics, Turn the Lens Jeff Frick, Future of Humanoid Robots.

#Robotics #AI #GoogleDeepMind #WorldModels #MachineLearning #TechTrends #EmbodiedAI #DataScience #Innovation #JeffFrick #TurnTheLens #FutureTech #HumanoidsSummit