What if the biggest bottleneck in robotics is the way robots are taught?

Jagdeep Singh believes the industry needs a much larger and more diverse source of training data to build robots that can handle the unpredictability of real work. At Rhoda AI, that starts with internet video.

In this episode of Automated, Brian Heater speaks with Jagdeep, co-founder and CEO of Rhoda AI, about the company’s approach to physical AI, why he sees limits in vision-language-action models, and what it will take to move generalist robots from demonstrations into factories and warehouses.

Jagdeep explains how Rhoda pretrains its model on hundreds of millions of video clips to learn how the physical world moves. The goal is to build an understanding of motion, forces, and object behavior before teaching a particular job. He says that foundation has allowed Rhoda to learn specific tasks with 10–20 hours of robot data, rather than the much larger datasets associated with other approaches.

The conversation explores why footage of a specific task is not always the best starting point. Jagdeep describes how broader video datasets performed better than more narrowly curated collections of people doing jobs. Task-specific learning comes afterward, followed by additional learning from the unusual situations robots encounter with customers.

That distinction became clear when Rhoda took a container-folding task from the lab into a factory. Broken latches and rusted equipment changed the conditions the model had learned. Jagdeep says the team adapted using roughly three hours of additional factory data, illustrating why deployment requires more than repeating a successful demonstration.

Brian and Jagdeep also discuss how learning can compound across customers, why access to deployment data is central to Rhoda’s business, and the difference between customers owning their data and allowing the company to use it for model training.

The conversation then turns to hardware. Jagdeep outlines four requirements driving Rhoda’s development: payload capacity, reliability, safety and dexterity. He explains why the company is building components it could not buy, how it is weighing the tradeoffs of three-finger hands, and why its mobile manipulators use wheels.

Finally, Jagdeep explains why customers ultimately judge robots by success rate and cycle time. He shares Rhoda’s plans to deploy its model on existing hardware while developing its own platform, the role of human demonstrations in teaching new tasks, and why reliable performance in customer environments is the test that matters.

Connect with Jagdeep Singh
https://www.linkedin.com/in/jagdeep-singh-3b2a9222

Learn more about Rhoda AI
https://www.rhoda.ai/

Explore Rhoda AI’s research
https://www.rhoda.ai/research/direct-video-action

We’d love to hear from you. Have thoughts or guest suggestions?
Reach us at podcast@automate.org.

You can find the transcript and more episodes of Automated at automated.fm.

Unlock full access to Automated and explore everything automation.
Subscribe today and leave a review on YouTube, Apple Podcasts, Spotify, and the Automated Newsletter.

https://www.youtube.com/@automatedpodcast
https://podcasts.apple.com/us/podcast/automated-with-brian-heater/id1837762221
https://open.spotify.com/show/60olq6brlBEIJWggx2fMR6
https://www.automate.org/automation/automated-newsletter

LinkedIn
https://www.linkedin.com/showcase/automated-podcast-by-a3/

Instagram
https://www.instagram.com/automatedpod/

Jagdeep Singh: Teaching Robots From Internet Video at Rhoda AI

Automated with Brian Heater

Jagdeep Singh: Teaching Robots From Internet Video at Rhoda AI