What the work actually is

You watch short video sessions of a robotic arm attempting an assigned task — grasping, placing, opening, inserting, retrying after a slip — and you convert what you see into structured labels. That means identifying the discrete actions inside a session, marking where each one begins and ends to frame-level precision, tagging the outcome according to the project's grading guidelines, and flagging cases the rubric doesn't cleanly cover. A large share of the job is boundary judgment: deciding the exact frame a grasp becomes a grasp, whether a fumble-then-recover counts as one attempt or two, and whether an apparent success is contaminated by an off-camera assist. Drift between your timestamps and the footage is treated as a defect, not a rounding error, so self-verification passes are part of the expected workflow.

What the screen is looking for

micro1's screening is AI-led and follow-up heavy. It probes whether you have genuinely used timeline-based annotation tooling — frame rates, timecode versus wall-clock, keyframe stepping, how you handle a video whose reported FPS doesn't match reality — rather than whether you can describe attention to detail. Expect scenarios: an ambiguous rubric clause, two annotators disagreeing on a boundary, a batch you've already submitted that a QA review says has a systematic offset. Prior robotics, ML data-preparation, or video-annotation experience is strongly preferred and is where most candidates separate; if you have it, be ready to name the tool, the dataset scale, and the specific label schema you worked under.

Logistics

  • Fully remote, contractor engagement, asynchronous within milestone deadlines.
  • Hours are typically flexible but project-batched — availability commitments matter more than a fixed schedule.
  • Requires a machine and connection that can scrub high-resolution video smoothly; stuttering playback makes frame-accurate work impossible.
  • Ongoing quality reviews mean feedback cycles: your early batches will be audited and you're expected to absorb corrections into later work.
  • Pay band of $50–90/hr reflects rates observed on this listing, not a guarantee; actual rate depends on the project and assessed level.