Skip to content
Koidra
Book a demo

Step 4 of five

KoPilot

Closed-loop control of the process, minute by minute, with no human at the panel.

The closed loop, and the envelope it obeys The process is read by a physics-aware model, whose proposed setpoints pass through an operating envelope that states the limits it may not cross, then reach the controllers, which write them back to the process. The loop runs every minute. The process dryer, boiler, hall, crop Physics-aware model knows what the equipment can and cannot do Operating envelope the limits it may not cross Controllers setpoints written back every minute, not every shift
The envelope is a separate box on purpose. It states what the loop may not do, and it is engineered for the site rather than learned from it, which is why a model that has never seen your plant is still safe to run on it.

Replaces

  • Setpoints tuned by hand, differently on every shift
  • Comfort margins nobody has revisited

Pays for itself by

Energy per tonne, yield, and consistency that no longer depend on who is on shift.

The number a buyer checks: Energy cost per unit of output, and output variance between shifts.

What KoPilot does

It holds the process at the operating point that costs least, and it does that continuously rather than whenever someone gets to it. It reads the plant through the lower levels, decides the setpoints, and writes them back to the controllers.

It has already beaten expert humans, twice

Koidra won the international Autonomous Greenhouse Challenge in two separate competitions, in 2018 and again in 2022. Not a simulation and not a pilot study: a real crop in a real facility over a full season, scored against teams of expert operators, with nobody at the panel.

The 2018 run finished 6% ahead on yield and 17% ahead on net profit against the manual reference the expert growers operated, and it was the only AI entry that beat them.

Competition is a harder test than commerce, because everyone is trying. In commercial greenhouse operations the numbers are larger: up to 27% yield improvement and 17% energy savings.

This is the only reason we are willing to use the word autonomous without qualifying it. An AI that ran a commercial operation better than the people who do it for a living is a different claim from software that suggests a setpoint.

Physics-aware, which is what makes it safe

Our models are grounded in how heat, mass and machinery behave, rather than fitted to last quarter’s numbers. The model knows a dryer cannot heat faster than its mass allows, so it never proposes it. It knows what a screen movement does to humidity two hours later.

Statistical models learn what usually happened. That is fine for a dashboard and dangerous for a control loop, because the situations worth automating are the ones that have not happened often.

Start with a conversation

Whether closed-loop control fits your plant depends on things a web page cannot know: what you instrument, how well the physics of your process is understood, and whether there is somewhere a supervised trial can run before anything operates unattended.

Most customers start at the first three levels, and that is usually the right order. Seeing the plant and running the work well are worth doing whether or not you ever hand over the setpoints, and they are what makes the handover safe when you do.

Tell us about your process and we will give you an honest read on what is feasible, in what order, and what it would take.

Where this sits

Four steps to autonomous operations. Each one is a complete product with its own payback, and each makes the next possible. Read the full ladder.

  1. STEP 1

    Essentials

    Every signal the plant produces, in one place, live and historical.

  2. STEP 2

    Maintenance Management

    The system decides what maintenance is due, dispatches it, and tracks what it cost.

  3. STEP 3

    AI Advisory

    Reports that arrive without being asked for, saying what changed and what to look at.

  4. STEP 4

    KoPilot

    Closed-loop control of the process, minute by minute, with no human at the panel.

See Koidra on your own plant data

Bring a week of readings from one line. We will show you what the platform sees, what it would have flagged, and what it would have scheduled.

Book a demo