Data Annotation Specialist

About This Role

The Data Annotation Specialist labels and quality-checks robot episode data, which is used for training and evaluating manipulation models. This role involves using internal annotation tools to review recordings, apply labeling guidelines consistently, and flag edge cases for researcher review, directly impacting model performance.

Responsibilities

• Annotate robot episodes (video + metadata) using internal tooling, following labeling guidelines precisely
• Maintain high quality standards by performing self-QC, spot checks, and fixing labeling errors before submission
• Apply consistent judgment across large batches of data and document ambiguous cases
• Track progress and output (batch completion, rework rate, error categories) and meet daily/weekly throughput goals
• Handle data securely and follow confidentiality and access-control procedures
• Flag edge cases or unclear labeling scenarios; work with researchers/ops leads to resolve ambiguities and improve guidelines
• Contribute to guideline iteration by suggesting clearer definitions, examples, and 'golden set' references
• Support dataset audits (validating that metadata is complete, identifying corrupted/missing files, noting systematic issues)
• Provide feedback on annotation tooling and workflow improvements to increase accuracy and throughput

Requirements

• Strong attention to detail and ability to apply rules consistently over long periods
• Comfortable learning new software quickly and working in computer-based annotation tools all day
• Clear written communication (able to document issues and ambiguity cleanly)
• Reliable attendance and ability to work full-time on-site
• Ability to maintain confidentiality and follow security procedures

Nice to Have

• Prior experience with data labeling/annotation (video, image, or time-series)
• Familiarity with basic QA concepts (spot checks, rework loops, 'golden sets,' inter-annotator agreement)
• Comfort with basic technical concepts (file systems, naming conventions, spreadsheets, simple debugging of tool issues)
• Interest in robotics/AI workflows and willingness to learn domain-specific labeling taxonomies

Benefits