Physical AI research
Models that learn how the physical world behaves.
Movendi builds the models. Vitesse puts them into live industrial operations and brings back what the world actually does. Every turn of that loop makes the next model better.
The idea
Most models learn the world by watching it. Ours learn it by running it.
Almost all the physical data available to train on is internet video and simulation — the world recorded, or the world approximated. Movendi trains on the world running: machines under load, in operations that cannot afford to be wrong.
How it works together
01 / Research
Movendi builds the models
Reality Models: AI systems that learn to predict how the physical world evolves, from observation rather than from engineered rules.
02 / Operations
Vitesse deploys them
The digitalization solution for industrial operations. It connects the equipment already on the floor, digitizes how the plant runs, and carries the models to real machines.
03 / Data
Reality comes back
Every deployment records what actually happened — real equipment, real variance, real failure. That record returns to the research as training material.
Why it compounds
Vitesse earns its place in the plant on its own: it digitizes operations that were dark, and customers pay for that. What it returns is training data no one can buy. Better models make the platform worth more. A platform worth more reaches more operations. More operations make the models better.
A lab without operations has no reality to learn from. A platform without research has nothing new to deploy.
The scale of the feedstock
800,000 machine hours
Monitored across customer sites today. Real equipment, under real load.
55 gateways
In production across manufacturing sites in North America and Europe.
135 native connectors
Into the PLCs, CNCs, and robots already on the floor.
A press controller from 1998.
A Siemens S7-300 at a European forging plant, previously unreachable. Connected in under an hour, streaming to AI models the same day. On another S7: 48,454 symbols discovered in seconds.
What is real, and what is not yet
Today
Deployment is real. Vitesse runs in production, digitizing operations on equipment that was never designed to be read.
Today
Perception and actuation are real. Edge computer vision from Kaliber Labs, retrained for the floor, on a control stack proven in production.
Building
Reality Models are the frontier. One learned system that predicts how the world evolves, in place of the stack we assemble today.