DeepRacer Resources

This is the place to find all the necessary resources for DeepRacer and our local Santa Cruz Racing League.

What is DeepRacer

DeepRacer is a 1/18th scale autonomous driving car powered by computer vision and machine learning. To start you build a machine learning “model” in the AWS cloud using the DeepRacer Console. Only basic coding knowledge is required to modify the sample “reward” code.

Once your model has trained, you can then transfer the model to a DeepRacer car and test your model’s ability to drive autonomously around a track. The car drives autonomously by using Computer Vision, using the video frames received to determine the track location and next action (i.e. which way to turn the wheels).

You can read more on the AWS site here

Getting Started

DeepRacer was created to make Machine Learning approachable. Don’t be afraid, it’s pretty easy to get going. The first thing you need to do is create an AWS account. The account is free (but credit card required) and includes a “free tier” which will cover some of your DeepRacer training time. Sign up here.

Next, watch the video. Don’t worry about understanding everything, I promise it will all sink in over time. Please note the console has changed since this video was produced, but the principals are the same.

Jump into your DeepRacer console and follow the Getting Started info. Next you will :

  • Choose (or build) a car in the garage
  • Create and train your model in the console
  • Evaluate your model

Our recommendation is to build a very simple car and train with the defaults. This allows you create a working model and understand the process. From there you can tweak the reward function, hyperparameters, and your car’s attributes. Specifically you should :

  • Use the The Original DeepRacer car
  • Choose the re:Invent 2018 track
  • Choose the Time Trial race type
  • Chose the “Time trial – prevent zig-zag” reward function (not the default)
  • Train for no more that 60 minutes

Community Resources

  1. Join the DeepRacing Global Community online. There is a Slack ChannelBlog, and Wiki. These are fabulous resources for you to learn and get answers on a near 24 hour basis. More at https://deepracing.io/
  2. Join our DeepRace Meetup Group
  3. Read blog posts on AWS (category and tag), and Medium (and Google)
  4. Watch DeepRacer TV on YouTube or Amazon Prime!
  5. Take the free AWS DeepRacer Training Course
  6. Our local Meetup Message Board