Vision that learns from change

Imagine AI vision systems that adapt as conditions change, whether on a field robot, in a glasshouse, or warehouse or looking at grain. Synthetic worlds. Precisely labelled training data. Available when you need it.

ARGOS

ARGOS watches over agricultural AI systems and helps them adapt when the real world changes.

When a robot encounters new crops, weeds or changing field conditions, its vision system must keep pace. ARGOS connects real-world observations with synthetic training data to help improve, test and update AI models rapidly, reducing the need for lengthy manual data collection and retraining.

SEE WHAT YOU CAN'T SEE

A crop that never existed

This field scene looks real, but it isn't. In the background, it is completely synthetic... modelled in software.

SYNTHETIC FIELD IMAGE — PLACEHOLDER

SEE WHAT WE CAN'T EAT

This fruit will never be eaten

Because they aren't real. They never existed. But the AI model doesn't know that because it lives in the same synthetic world as the fruit.

VISION CHANGES

The world changes every second.
Can your vision system?

Your system stops because it can't classify the objects it's seeing. Maybe the light changed. Maybe it's a contaminant, maybe a new weed or disease. What do you feel? What do you tell the customer? How long will it have to pause before we can retrain? Now we have a better way forward.

ADAPTIVE VISION DEMO — PLACEHOLDER

APPLICATION VIDEO — PLACEHOLDER

APPLICATIONS

From field robots to conveyors

Although the use cases vary tremendously, the principle is the same. Something changes.
In the field, it can be tin cans, dead rabbits, autumn fallen leaves, a new soil type. In the glasshouse, it can be a disease, a new variety, a lighting change, a new pot filling. In a warehouse, it can be moisture on a fruit. Throughout the whole journey of our food from the field or glasshouse to consumption, the scene can change.

What is the synthetic data that drives ARGOS?

We create models of grain, plants, fruits and other objects like crop debris (trash). We manipulate those models in a scene with other objects to deliver thousands of images, each one unique and each part of a precisely annotated dataset with controlled scope and variance. Every image is different, and each precisely annotated dataset has a controlled scope and variance, normally just beyond what you would consider Edge Cases.