Training on the equipment meant having an expert standing next to you.

Detectors can be bought. People qualified to stand behind a trainee and say “you swung too fast there, you lifted the head, you walked past that patch” cannot. Every hour of training needs one of them present, which puts a hard ceiling on how many people can be brought up to standard and how quickly.

It also makes the feedback inconsistent. Two instructors watching the same sweep will not always call it the same way, and neither can replay what actually happened.

A tablet application that reads the detector while it is in use and reconstructs where the head actually went, in real space, as the trainee sweeps. It shows the ground they covered and the ground they did not: where they moved too fast, where they lifted the head out of range, what they walked past.

The engineering problem is that none of the four inputs is sufficient alone and none of them is exactly right. Satellite positioning knows where the operator is, not where the head is. The motion sensors know how the head moved but drift over time. Reconciling them into one position, continuously, accurately enough that a trainee should believe the replay, is the whole job.

The application reports what happened during a sweep and shows it back. Judging whether that sweep was good enough, and what the trainee should do differently, stayed with the instructor. It was built to make an expert’s attention repeatable and reviewable, not to replace their judgement about a person’s readiness.

That distinction mattered more here than anywhere else we have worked. The consequence of a system quietly deciding someone was competent is not a bad customer experience.

Red Lotus Technologies. We delivered it as their engineering team, four people working directly with their chief technology officer, and the product shipped under their name into service with the US Army. Red Lotus has since wound down.

How we connect systems →