Organising a DeepRacer Event

From AWS DeepRacer Community Wiki
Revision as of 07:02, 1 December 2019 by Tptak (talk | contribs)
Jump to navigation Jump to search

It's great that you want to organise an AWS DeepRacer event. Depending on your circumstances it can take a number of forms.

Key elements for an event:

  • goal
  • target group
  • venue
  • AWS DeepRacer
  • a track

Let's discuss them slightly more.


Target group


AWS DeepRacer

A track

The track is usually the most tricky part of preparing for an event - if you want physical racing to happen you need the space (re:Invent 2018: 5x8m, re:Invent 2019: 10x8m) and the track itself. Read more in Preparing a racing track.


Anya would like to create a DeepRacer group at work. She has taken part in AWS DeepRacer Virtual League and liked it, and has learned a lot. She believes her team would also like it and the company would benefit from new knowledge. She doesn't have a car or a track.

Goal: education, team building Target group: colleagues at the company Venue: company office AWS DeepRacer: None Track: None

What Anya could try doing is to present AWS DeepRacer at a tech chat. The presentation could describe:

  • the car
  • the league
  • machine learning and where reinforcement learning fits in
  • how AWS DeepRacer learns
  • what other skills one can obtain
  • her experience with AWS DeepRacer
  • encouragement to build a racing team at the company

Based on the output, Anya can follow up with a workshop to help others train their first model and, if a race is ongoing, submit it for the competition. Such workshop could describe:

  • AWS DeepRacer Console
  • reward function
  • PPO algorithm
  • hyperparameters
  • AWS Sagemaker - mention where the training job is happening
  • AWS Robomaker - show the simulation job and what elements one can view
  • CloudWatch - show where the logs are
  • S3 - the layout of the bucket

One thing that should be considered is whether the company would allow for employees to train using company accounts. If yes, certain permissions need to be granted. Optionally separate accounts can be created and disposed of after the workshop: read the documentation. If not, users need to have free tier accounts. Do warn about the costs.

If the racing continues, other topics could include: log analysis, training using SageMaker notebook example, advanced training techniques.

You can see that a lot can be made without a car or the track. Depending on company involvement and interest it may be possible to obtain a car and a track or invite an organisation that has those and offers a day racing, for instance AWS.

External resourced

Article about how OpenMarket is using DeepRacer to promote machine learning