Master AnnotatorMaster Annotator

Introduction

Master Annotator is a web-based platform for labeling and visualizing robotic motion data. Turn raw trajectory recordings into structured, labeled datasets ready for machine learning.

What is Master Annotator?

Master Annotator provides a complete workflow for annotating robot trajectories. You upload CSV files containing joint angles and end-effector positions, configure your robot model for 3D visualization, define a label schema, and then annotate each frame of the trajectory with semantic labels.

The platform supports multiple robot types, real-time 3D visualization, synchronized video playback, and flexible export formats — all from your browser with no installation required.

Key Features

  • 3D Robot Visualization — See your robot move through trajectories in real-time with support for generic articulated arms and URDF-based models (KUKA, Franka, Kinova, UR).
  • Multiple Labeling Modes — Single-frame toggling, auto-advance, range painting, and keyframe interpolation to match your annotation workflow.
  • Video Synchronization — Pair MP4 recordings with CSV data for side-by-side verification of robot behavior.
  • Flexible Export — Export labeled data as CSV with configurable column layouts, delimiters, and value mappings.
  • Label Templates — Save and reuse label schemas across projects for consistent annotation vocabularies.
  • Joint Angle Graphs — Visualize joint trajectories as interactive charts synchronized with the 3D viewer and timeline.
  • Keyboard-Driven Workflow — Assign custom keyboard shortcuts to labels for rapid annotation without mouse clicks.

Who is it for?

Master Annotator is built for:

  • Robotics researchers who need to segment and label teleoperation recordings for imitation learning.
  • Data engineers preparing training datasets from robot demonstrations.
  • ML teams that need labeled action primitives (reach, grasp, lift, pour, place) from continuous trajectories.
  • Lab managers coordinating annotation work across team members with shared projects and organizations.

How it works

The platform follows a simple hierarchy:

  1. Create an Organization — Your top-level workspace for grouping projects.
  2. Create a Project — Each project has a robot configuration, a label schema, and one or more datasets.
  3. Upload Datasets — Upload CSV files (and optional MP4 video) containing your robot trajectory data.
  4. Configure Your Robot — Set up the robot model so the 3D viewer can render your trajectories accurately.
  5. Define Labels — Create the annotation vocabulary with colors, keyboard shortcuts, and export values.
  6. Annotate Frames — Use the labeling interface to apply labels across your trajectory frames.
  7. Export — Download your labeled data as a CSV file ready for training pipelines.

Next steps

  • Follow the Quickstart guide to label your first dataset in under 5 minutes.
  • Read Concepts to understand the domain hierarchy in detail.
  • Check CSV Format for data preparation requirements.
Quickstart