Measuring what machines do around people

Software used to be checked with tests that pass the same way every time. Trained systems are different. They are reliable rather than repeatable: most of the time they do the right thing, and sometimes they don't.

Once those systems move into the physical world, into factories, care settings and homes, a mistake is no longer a wrong answer on a screen. It is a dropped load, a hard contact or a fall. That kind of harm can be measured, insured and regulated, and it will need to be.

We already know how to manage capable but fallible actors: training, licences and independent checks. Physical AI will need the same, and every one of those depends on measurement that someone else can verify.

Benchmarks show what a system can do with its tools. Independent, traceable measurement shows what it does under defined conditions.

A testing laboratory that measures force, torque, contact and friction for robots and the people who work with them: an independent reference that makers, deployers and insurers can check claims against.

• Results in SI units, each with a declared uncertainty.
• Designed to be traceable to Canada's national measurement standards.
• Independent: we do not build robots, we do not sell data, and we do not choose which AI anyone uses. We measure.

From 20 January 2027, EU rules require independent conformity assessment for machinery whose safety functions rely on machine learning. Makers everywhere will need test evidence they can stand behind. Independent measurement for learning machines is still an early field, and the methods are being written now.

• Physical behaviour, not cognition. We measure what machines physically do around people. We never decide who may use AI, or which model, open or closed, from any country, a maker chooses.
• Measure, don't sell data. We provide measurements and test results, not datasets.
• Show the uncertainty. Every result states how sure it is.
• Open criteria, no monopoly. Our aim is that the criteria we test against are published, and that other qualified laboratories can test against the same references.
• Published fees. Our aim is that our fees are published and the same for comparable work: the cost of the work plus a fair return that we disclose, never set by market power and never dependent on the result.
• Results do not depend on the outcome. What a client hopes to hear has no bearing on what we report.
• Consent, fair share and new skilled work. Anyone whose movement is measured gives informed consent, and that record stays attached to what was measured. Our aim is that the people who contribute share fairly in what their contribution helps build, and that the laboratory creates new skilled work in measurement.
• Keep the tester separate from the tested. Independence is designed in from the start.

Early. The laboratory is in development and working toward ISO/IEC 17025 accreditation, the international standard for testing laboratories. We do not issue certificates.

We would like to hear from:

• Researchers, including graduate students and postdocs, working in measurement, robotics, human–robot interaction or functional safety.
• Companies developing robots or wearable systems that need independent characterisation of force, contact or grip.
• People working on standards for robotics and AI safety.