What the work actually is

You watch recorded sessions of robotic arms — pick-and-place, assembly, insertion, bin-picking, and similar industrial routines — and tag each one according to a grading guideline the customer provides. That means classifying the outcome, marking the moment things go wrong, and writing a short technical comment on the cause: a grasp that slipped because of approach angle, an overshoot from aggressive gains, a collision from bad path planning, compliance that never engaged, a trajectory that hit a singularity. The value you add over a generalist annotator is diagnostic: you can tell a controller problem from a perception problem from a fixturing problem by watching the motion.

Expect the rubric to evolve. Early batches surface edge cases nobody anticipated, and you'll be asked to flag ambiguity rather than silently invent a rule. Consistency across your own annotations, and agreement with other annotators on the same clips, is the metric that matters most — a brilliant but idiosyncratic label set is worse for model training than a plainer one applied the same way every time.

What the screen looks for

micro1 runs an AI-led interview. It probes for a real mechatronics background — degree or equivalent working history — and then pushes on failure diagnosis: given a described motion, what are the candidate causes and how would you distinguish them from video alone. It also tests annotation discipline: what you do when the guideline doesn't cover a case, how you keep labelling consistent over hundreds of clips, and whether you'll escalate rather than guess. Prior data-labelling or AI-training exposure helps but is not the gate; demonstrable ability to read machine motion is.

Logistics

  • Contractor engagement, fully remote, asynchronous — no fixed shift, but batch deadlines are real
  • Open to candidates in Northern America, Latin America, Europe, the UK, and Oceania
  • Observed band of $50–90/hr; actual rate depends on batch, seniority, and region and is not guaranteed
  • Requires a machine and connection that can stream and scrub video reliably, often frame by frame
  • Volume is project-driven and can be uneven; treat it as a supplement rather than a guaranteed weekly load