Vision-Language-Action (VLA) Annotator

About the role

A remote, full-time or contract position for detail-oriented VLA Annotators who help build high-quality training data for Vision-Language-Action and robotics models. Annotators watch videos of people doing real-world activities, split them into action segments and write or check clear descriptions of what happens. Projects cover household, workplace, industrial, manipulation, navigation and other activities.

Key responsibilities

• Review videos and understand the overall task or activity being performed.
• Divide videos into logical action-based segments with accurate start and end times, capturing all meaningful actions without gaps or overlaps.
• Review AI-generated captions and select, edit or replace them when necessary.
• Write clear, concise action descriptions based only on what is visible, describing hands, objects, tools, movements and locations accurately.
• Maintain consistent terminology for objects and actions throughout each annotation.
• Identify invalid, idle, distracted, failed, recovery or otherwise unusual portions of a video per project guidelines.
• Review completed annotations for segmentation, caption accuracy, grammar, consistency and completeness before submission.
• Follow project-specific guidelines, flag ambiguous examples, model errors and edge cases to the Operations/Quality team, and use annotation shortcuts and AI-assisted tools efficiently.
• Take part in calibration, training and feedback sessions as needed.

Required profile

• Strong attention to detail and ability to follow detailed written guidelines.
• Good written English and strong observational skills to spot small differences in actions, objects and timing.
• Comfortable reviewing video for long periods and able to do repetitive work with consistent accuracy.
• Basic computer proficiency, ability to learn new platforms quickly, and a reliable computer and internet connection.
• Nice to have: previous data or video annotation, QA, robotics data or computer vision experience; familiarity with temporal segmentation, egocentric video or AI-assisted labeling tools.