What the work actually involves

You will spend most of your time inside a video annotation tool, watching footage captured by robot platforms — manipulators, mobile bases, or similar — and marking what is happening in it. That means drawing and tracking boxes or masks on objects, segmenting a clip into events (grasp attempt begins, contact, slip, recovery), and flagging anomalies such as occlusion, motion blur, sensor dropouts, or a gripper that closes on nothing. Alongside the labels, you write up what you saw: ambiguous frames, categories the taxonomy does not cover cleanly, and places where two annotators would plausibly disagree.

The second half of the job is protocol work. Annotation guidelines on robotics datasets are rarely finished when you arrive. You will be asked to sit in periodic review rounds, defend or revise your calls on disputed clips, and propose edits to the labeling spec when the edge cases pile up. Suggestions about data capture itself — camera placement, frame rate, lighting conditions that make a class unlabelable — are explicitly in scope and tend to be the highest-value thing a specialist contributes.

What the screen looks for

micro1's process is AI-led: a structured voice interview followed by review. The listing says no AI experience is required, and that is honest — nobody will ask you to train a model. What is probed is whether you have actually done video-level vision work rather than still-image classification, whether you can reason precisely about temporal boundaries and occlusion, and whether you can articulate a labeling decision well enough that someone else could reproduce it. Vague claims collapse fast under follow-up; specific datasets, tools, and disagreement stories hold up.

Logistics

  • Fully remote, contractor engagement, hourly billing
  • Largely async annotation work with scheduled review sessions; expect some overlap with the client team's hours
  • Volume varies by project phase — treat it as part-time-flexible unless told otherwise
  • Observed band is $50–90/hr; rates are set per engagement and not guaranteed