Moving Atoms (YC S26)

World Models for Robots.
Robots don't have enough data. So we're making foundation world models for them.

[Moving Atoms (YC S26) @MovingAtomsLab]

@SaroyaShehryar and @AhmadSaroya00 are brothers who both studied maths at Oxford.
Shehryar designed, trained, and deployed the core models for TikTok’s Video Recommendation AI. Ahmad scored top of his college at Oxford while he was studying there.

[Moving Atoms (YC S26) @MovingAtomsLab]

We're solving robotics with world models trained on internet-scale video: models that accurately predict physical interactions and are conditioned on robot actions.
Our first model, Atom 1, already tops DeepMind’s Physics IQ benchmark - beating NVIDIA's Cosmos 3.

[Moving Atoms (YC S26) @MovingAtomsLab]

LLMs were unlocked by the trillions of text tokens on the internet. Moving Atoms believe robotics gets unlocked a similar way, by the latent physics information in online video content.