The Blanket

Small world models for real factories

Robots that learn
while they act.

No code. No cloud. No datasets.

Our product

auto·motion.

A model small enough to run on the robot, and smart enough to adapt.

auto·motion learns a task from just a few demonstrations by building a compact world model that runs directly on the robot. The longer it runs, the better it gets, and your data never leaves your premises.

One task, learned fast
  1. You

    Demonstrate

    Show the task a few times. That’s your part.

  2. Autonomous

    Model update

    Builds a compact world model of the task.

  3. Autonomous

    Run on the edge

    On the robot. Data stays on your premises.

  4. Autonomous

    Adaptive execution

    Watches for surprises while executing: looks again when unsure, and recovers when something slips.

  5. Autonomous

    Improve

    Every run makes the model better.

Every run feeds back into the model update
Why us

auto·motion is different by design.

Hand-programmed robots break when a part moves. Data-driven AI needs huge datasets and data centres. In contrast, auto·motion learns from a few demonstrations, adapts on the job, and runs on the robot itself.

A programmed robot needs no data, but someone has to write its code. Data-driven AI needs large datasets. auto·motion needs a few demonstrations.

A programmed robot runs on its controller and keeps its data on site. Data-driven AI usually needs the cloud or data-centre GPUs, so your data often leaves the building. auto·motion runs on the robot’s own computer: everything it learns stays on the robot and never leaves your premises.

A programmed robot has to be reprogrammed for every variation. Data-driven AI handles variations too, but only the ones its dataset covers: each new variation means collecting more data and retraining. auto·motion adapts online, on the job, from what it sees. And when it is unsure, it says so, looks again, and recovers, where the others carry on regardless.

A programmed robot repeats one scripted motion; anything more complex means more code. Data-driven AI handles complex, varied tasks, given enough data and compute. auto·motion handles complex tasks too, learning them from a few demonstrations by building a world model of the task.

Click a row for more.

About

Why “The Blanket”?

We’re named after the Markov blanket: the boundary between an agent and its world, through which it senses and acts. We believe adaptive intelligence comes from the continuous loop of understanding and acting.

Active inference, a framework from neuroscience for how living systems learn and act under uncertainty, describes exactly that loop. It is the science behind auto·motion.

What problems would you solve with a robot you can teach in minutes?

contact@theblanket.ai LinkedIn