FTC Machine Learning Toolchain

This tool, the FIRST Tech Challenge Machine Learning toolchain (FTC-ML), allows FIRST Tech Challenge teams to create custom TensorFlow models for use in the game challenge. Learn how to train TensorFlow to recognize your Team Shipping Element, Freight scoring elements, and more using this tool, and download models that you can use in your autonomous and driver-controlled Op Modes.

• Machine Learning In a Nutshell
• Logging in to FIRST Tech Challenge Machine Learning
  • Logging into the ftc-ml tool
  • Changing the active team login session
  • Logging out
  • Adding students to your team’s ftc-ml workspace
• Managing the ftc-ml tool workflow
  • Overview
  • Creating videos of objects to be recognized
  • Uploading videos to the ftc-ml tool
  • Adding labels to frames in a video
  • Producing Datasets
  • Training Models
  • Continuing Training on Models
  • Understanding Model Metrics
  • Canceling Training
  • Deleting a Model
  • Downloading Models
• Implementing in Robot Code
  • Android Studio
  • OnBot Java (OBJ)
  • Blocks
• Optimizing Videos for increased TensorFlow Model Performance
• FAQ
  • Why is TensorFlow called “TensorFlow”?
  • How many frames of our object is enough to ensure a good model?
  • How do I know if my model is trained well?
  • Why does my team only get 240 minutes of model training time?
  • Why can’t I seem to get a 100% object detection prediction?
  • I read somewhere about a parameter I can tweak…
  • Can object bounding boxes overlap?
  • What are the limitations imposed within the ftc-ml tool? (PER TEAM)

Volunteer Special Thanks

The FIRST Tech Challenge staff would like to extend a special thanks to the following volunteers for their hard work and dedication toward this project:

• Liz Looney, Google – FIRST Machine Learning Toolchain lead developer

• Mr. Phil Malone – Model designer and platform tester

• Uday Vidyadharan, Team 7350 – Platform tester and Contributor

• Jacob Burroughs – Platform configuration and FTC Scoring SSO

• Richard Lester – Platform UI improvements