Skip to content
Robotoplus

Vision

Inspection and monitoring packages built on the platform's own sensor stack.

2 products

Research projects, and what that word is doing here

Both items in this category are research projects. That is not a disclaimer added for safety, it is the commercial arrangement: they are billed on the engineering days actually spent rather than against a fixed scope, because neither problem has a known answer on your data until someone has looked at your data.

What that buys is a working system on your own material
, built and tuned in front of you, with the option to stop when the answer is good enough or when it becomes clear it will not be. What it does not buy is a fixed price for a fixed result. A supplier offering one for this class of work is either doing something simpler than it sounds or absorbing a risk they have not mentioned.
The language model behind both is yours, not ours.
You bring the account, whether that is OpenAI, another cloud provider, or a local deployment such as Ollama on your own hardware, and the usage is billed to you directly. We take no margin on tokens and we are not reselling anyone's model. The architecture is deliberately provider-agnostic, so the choice stays a decision you can revisit rather than a dependency you inherit from us.
For the monitoring work in particular, that matters more than it sounds.
A model running locally on your own hardware means footage of your production floor never leaves your building, and for a system whose whole job is watching people work, that is usually what decides whether it can be deployed at all.

What the system decides, and what a person still has to

Neither of these is a safety function, and that distinction is the most important thing on this page.

The monitoring package moves through a production area
, forms a view about whether what it sees looks safe or unsafe, and reports what it finds to your EHS officer. A person reads that report and decides what happens next. It does not stop a machine, it does not gate an interlock, and nobody's protection should depend on it having been right. We turn down work where a person's safety rests on a robot behaving correctly rather than on a guard or an interlock, and this is that same principle applied to a system that only ever writes a report.
The inspection package has the same shape.
It compares an incoming insulator against its technical drawing and flags visible deviations. That is a first pass which separates the obvious from the questionable, not a certified inspection, and it does not replace whatever your quality process signs today.
Three things have to come from you before either can start.
The data: drawings, photographs, documents, whatever the system is meant to reason against, in the state it is actually in rather than an idealised sample. The definition: what counts as unsafe, what counts as a defect, written down by someone who will still agree with it in six months. And a person who owns the output, because a report nobody reads is the most expensive kind of automation.
Agree the measurement before the work starts.
How many findings, checked against how many known cases, with false alarms and missed cases counted separately. A system that flags everything is as useless as one that flags nothing, and only the second one looks like it is failing.

Frequently asked questions

Contact us now