CosmicBrain AI claims one video can teach robots new skills instantly
“The more number of videos, the more autonomous the robots are.”
How much video does it take to train a new skill? So we're doing one to one transfer. So from a single episode to a single action. The more number of videos, the more autonomous the robots are.
Yeah. One to one is really fast.
That's crazy impressive. Yeah. I think Rohit's point was, like, Figure is selling a robot. Yeah. Why wouldn't they go to Amazon and say, hey, you know, give us some videos of, like, packages and we'll train the Figo robot. Like, why would why would you be required as a software piece in Google? So that's an excellent question. Right? So most of the robots,
they train, like, large, large foundation models, which require all this complexity. Second thing, it requests teleoperation to train these robots. Right? The humans have to wear exoskeletons,
About this clip
The founders explain their breakthrough approach to robot training that requires just one video demonstration per skill, contrasting with competitors like Figure who need massive datasets and human exoskeletons. They address the key question of why their software layer is necessary when robot manufacturers could theoretically train models themselves.
Why this clip
This clip reveals a significant technical differentiator and addresses a core business model challenge in the robotics AI space.
What they said next
How Chipotle's training data could teach every robot to make burritos
7:24 - 41s · market insight
More from this episode
Similar clips from other shows
From the blog
Want clips like this for your podcast?
We find your top 5-8 clips, write the hooks, and deliver ready-to-post content. First 2 episodes are free.
Get 2 Episodes Clipped Free