What the work actually is

You will spend your sessions watching footage of robotic arms attempting tasks — pick-and-place, insertion, handover, tool use — and labeling what happened against a written guideline document. That means segmenting the video into actions with defensible start and end boundaries, tagging events, and classifying outcomes: clean success, missed grasp, slip or drop, collision, poor approach path, inaccurate final pose, premature release. Where the robot fails, you are expected to say something about why — whether the failure looks like a perception problem, a control or trajectory issue, a gripper/force problem, or a fixture and integration issue. That diagnostic layer is what a general annotator cannot supply and what the pay band reflects.

Consistency matters more than cleverness here. The same borderline event seen in session 4 and session 400 must get the same label, and the guideline — not your personal engineering opinion — is the arbiter. Genuinely ambiguous cases get flagged to project managers rather than guessed at, and recurring ambiguity is a signal you're expected to surface as a proposed guideline amendment. Annotators who quietly invent a house rule to keep throughput up are the ones who damage a dataset.

What the screen looks for

  • Verifiable hands-on exposure to robotic arms — industrial cell, research lab, or integration work. Expect follow-ups on specific platforms, grippers, and what you were doing with them.
  • Diagnostic reading of motion from video alone. You will likely be asked to describe how you distinguish a grasp that failed from force control versus one that failed from a bad pose estimate, with no telemetry available.
  • Guideline discipline. Screens probe how you behave when the rules and your judgment disagree, and how you handle a case the rules never anticipated.
  • Tolerance for repetition. Long stretches of near-identical footage are the job. Honest answers about how you sustain accuracy beat claims of limitless patience.

Logistics

Fully remote contractor engagement, asynchronous, with annotation volume assigned in batches against deadlines and a quality bar. Hours are flexible in practice but the work rewards sustained blocks rather than scattered minutes — frame-level boundary marking is hard to do well in five-minute gaps. Expect a calibration or qualification round on sample footage before production work, and periodic quality audits against gold-standard labels. A stable connection and a display good enough to judge fine positional detail are practical requirements.