Master AnnotatorMaster Annotator

Quickstart

Get from raw trajectory data to your first labeled dataset in under 5 minutes. This guide walks through the complete workflow end-to-end.

Prerequisites

  • A Master Annotator account (free tier works for this guide)
  • A CSV file with robot trajectory data — one row per frame, with columns for joint angles (in radians)
  • Optionally, an MP4 video recording of the same trajectory for visual verification

1. Create an account

  1. Visit the homepage and click Get Started. You can sign up with your email or use Google OAuth. No credit card is required for the free tier.

2. Create an organization

  1. Open your dashboard after signing in, then click + New Organization and give it a name (e.g., "My Lab"). Organizations are the top-level container for all your projects.

3. Create a project

  1. Inside your organization, click + New Project. Provide a name and optional description. Each project is configured with:
    • Robot type — Choose from generic articulated arm, KUKA iiwa14, Franka Panda, Kinova J2S6S200, UR10, UR5 with gripper, or dVRK Classic PSM1.
    • Label schema — The set of labels you'll use to annotate (configured in a later step).

4. Upload data

  1. Navigate to the Datasets tab in your project. Click Upload Dataset and select your CSV file. The upload dialog will:
    1. Validate your CSV structure (headers, column count, file size)
    2. Count the total frames (rows)
    3. Detect timestamp columns for automatic FPS calculation (looks for timestamp, time, t, or elapsed_time columns)
    4. Optionally accept an MP4 video file to pair with the trajectory

    Your CSV must have at most 100 columns, column names under 50 characters, and be within your plan's file size limit (100 MB on Free, 300 MB on Standard, 500 MB on Pro & Enterprise).

5. Define labels

  1. Go to the Labels tab. Click + Add Label to create your annotation vocabulary. For each label, configure:
    • Name — A short descriptive name (e.g., "reach", "grasp", "lift")
    • Color — Pick from 10 color options for visual distinction
    • Keyboard shortcut — Assign a key (1-9, A-Z) for fast toggling
    • Export value — The string value used when exporting labeled data

    You can also load a saved template if you have reusable label sets from previous projects.

6. Start labeling

  1. Click on any dataset to open the labeling interface. You'll see:
    • 3D Robot Viewer — Real-time visualization of the robot at the current frame
    • Timeline — Scrub through frames with playback controls
    • Label Buttons — Toggle labels on the current frame with clicks or keyboard shortcuts
    • Joint Graphs — Line charts showing joint angle trajectories over time

    Press a number key (matching your label shortcut) to toggle a label on the current frame. Use ← / → arrow keys to step through frames. For faster annotation, switch to Auto-Advance mode which automatically moves to the next frame after labeling.

7. Export your data

  1. Click the Export button once you've labeled your frames. Choose your preferred layout:
    • Single column — All labels for a frame in one column, separated by a delimiter
    • Multi column — One boolean column per label

    The exported CSV preserves all your original data columns and appends the label columns. Download the file and use it directly in your training pipeline.

IntroductionConcepts