On the job
It learns while it works.
A brain can acquire useful skills through initial experience, then keep those skills while it acts and learns from further outcomes.
Pragma Research / Real-time AI
AI that learns from live experience.
Cadence is a novel real-time AI modeled after animal and human brains. It keeps learning while it works.
We build the brains for robots that keep improving throughout their working lives.
The Cadence core is open source. Get the code ↗
Cadence / A live brainCoupled models
Illustration
The film / 7 minutes
See how Cadence
thinks and learns.
The regions of the brain pass signals forward and back until the whole system agrees on one answer. Witnessed outcomes can teach the same brain; the goal is to preserve familiar skills and repair failed expectations.
The difference
Most AI stops learning
before it ships.
A typical deployed transformer keeps its learned weights fixed during use. Cadence explores a continuing equilibrium brain that can retain context and learn from actual outcomes.
Signals in Cadence loop through a reciprocal graph until its parts settle. Initial experience can build useful knowledge; familiar behavior should reuse it, while witnessed mismatch drives local repair. Measuring that selective repair and its cost is part of the research.
That matters most for machines with bodies. A robot's world does not hold still.
Made for a body
What a robot gains.
A robot meets wear, new loads and places nobody trained it for. A brain that learns from experience can keep up with them.
Changed conditions
It adapts when things change.
A worn motor, a heavier load or a different floor changes what the robot senses. The design goal is to repair the affected knowledge while retaining useful skills.
On the machine
It learns on the robot itself.
Connections can change from local neural activity, without a backward computation graph. Whether learning fits a robot's compute budget is measured on that body.
Shown in simulation today, in demos you can run yourself. A physical robot evaluation is the next test.
Working demonstrations
Watch a brain learn.
These three demonstrations start fresh in the browser and learn from their declared teachers or actual consequences.
Atari Arcade / Live learning
Watch brains learn to play.
Two brains are born when the page opens. They watch a teacher, take the controls and keep learning from every point they score.
Patch World / Learning and evolution
Watch brains learn and evolve.
Creatures learn in your browser from the consequences of their actions, and evolution reshapes their brains across generations.
Rover Lab / Adaptation
Change the body. Watch it adapt.
Weaken a simulated rover's wheel. The live brain relearns how its own body moves and keeps reaching its targets.
Build with us
We build the brains.
You build the bodies.
For robotics teams.
Bring one robot and one problem it struggles with. Together we test what a learning brain changes.
Explore a partnership ↗For investors.
Help take a working software brain from simulation into physical robots.
The investment case ↗Physics, mathematics and AI
Research behind
the brain.
Pragma Research is led by Bernhard Mueller. Cadence grew from Observer Patch Holography: research into how local observers build a shared account of a world.