I’ve been running the AWS DeepRacer Community for over a year now. In that time thousands of people have embarked on their machine learning journey, using the Slack channels as a way to close gaps in their knowledge. One of things I admire most is that many of these people then stay around in the Community and help the next generation of racers.
Occasionally, some of our Community decide to share their newly found expertise in more formal ways, such as blog posts and YouTube videos. I was recently approached by Daniel Gonzalez, a student who has been using DeepRacer for a class project. Daniel and his team wanted to share the findings of their rigorous experiments.
I have to say, this is one of the best write-ups & guides I’ve seen since the Community was founded. I’m impressed at the approach and even I learned new things from it!
You can read Daniel’s article on Towards Data Science.
In the article, Daniel presents the approach that got his university team to 12th place out of 1291 entries in the DeepRacer F1 Time Trial Event in May 2020. By sharing their approach, they hope it will help many people to develop their own improved strategies. I certainly think it will!
The article covers following topics:
1. A Short Introduction to AWS DeepRacer and their Setup
2. Computing the Optimal Racing Line and Speed
3. Optimizing the Action Space
4. The Reward Function
6. Continuous Improvement with Log Analysis
7. Automated Race Submissions with Selenium
8. Summary & Next Steps
Also, feel free to check out their GitHub Repo where they share some of their code.