What the work involves
You watch short video sessions of a robotic arm attempting a physical task — picking up an object, placing it, opening a drawer, pouring, stacking — and label what happens inside a dedicated annotation tool. That means setting boundaries on action segments, tagging object interactions and transitions, and recording whether the task succeeded, partially succeeded, or failed and why. The judgment being asked of you is the same one you apply when trimming a cut: deciding the exact frame where a motion becomes an action, where a grasp becomes a hold, where a slip becomes an error.
Most of the day is batch work against a customer-written guideline document, which is detailed and gets revised. You are expected to follow it literally rather than apply personal taste, and to flag rather than guess when footage is occluded, blurred, or genuinely ambiguous — with a short written note explaining what you could and could not see. Some batches involve reviewing other annotators' work and giving specific, guideline-referenced feedback.
What the screen looks for
- Verifiable editing or post-production background, and fluency talking about timecode, frame rates, and where you place a cut
- Whether you can hold consistency across hundreds of near-identical clips — this is the main failure mode on the project
- How you behave at ambiguity: escalating with a written observation versus quietly inventing a rule
- Willingness to accept a guideline that contradicts your instinct, and to ask about it in writing instead of overriding it
Logistics
Contractor, fully remote, asynchronous within batch deadlines. Open to candidates in North America, Latin America, Europe, the UK, and Oceania. Observed rates for this listing run $50–90/hr, which is what has been posted rather than a guaranteed offer; volume comes in batches and is not a fixed weekly commitment. You will need a reliable machine and connection capable of scrubbing video smoothly in a browser-based tool, plus a display good enough for frame-level review.