Video-based world models are rapidly displacing prior paradigms in robotics control.

A humanoid robot similar enough to a human can be trained on all available human video data, rather than requiring robot-specific embodied data.

“If you get the robot to be similar enough to a human then you can train on all of the available video data out there of.”

LTX Video's Dream Zero paper showed it is fairly easy to add joint encodings to video tokens to completely replace the previously dominant VA paradigm in robotics.

“Video showed in their dream zero paper that it's fairly easy to add to video tokens some kind of encoding of joints of the robot and then basically completely ditch the VA paradigm that was reigning supreme before it.”

Applied Intuition's neural simulation is a hybrid of Gaussian splatting and diffusion methods.

“We call our work in this neural simulation, but it's think of it like a hybrid of Gaussian splatting and diffusion methods.”

Robots currently require significantly more training data and examples than humans to learn tasks, contradicting the idea that AI robots learn faster than people.

“Robots are not very good at learning yet. Robots take so much more data, so many more examples than a person.”