PART-I
While most articles focus on virtual circuits, I’m here to simplify things for you and share tips that I’ve gathered from the helpful Discord community and other resources. These tips have helped me achieve an impressive top 8rank in the previous season and an outstanding top 5 position this season.
Welcome to this exciting blog post, where we explore the incredible world of AWS Deep Racer and provide valuable insights for students who are just starting out or those who want to improve their rank significantly. Get ready to dive in and discover the secrets of success in DeepRacer!
What is the difference between the Student League and Virtual Circuit?
There are two ways to get started with AWS DeepRacer: the Student League and Virtual Circuit.
Student League
The Student League is a free program designed for students who are just starting with machine learning. Here’s what you need to know:
No additional cost is involved for participating in it. You don’t need an AWS account or credit card to use AWS DeepRacer Student.
Every participant gets 10 hours of training for their reinforcement learning model, which is the main challenge in this competition.
It provides the course with access to 20 hours of foundational learning material designed by ML experts at AWS — for free.
Students who are underserved or underrepresented in tech can qualify to earn a Udacity nano degree scholarship through the AWS AI & ML Scholarship Program.
Student League Tips and Insights
We are limited in the student league and only given leverage to design our reward function, which is a craft that is difficult to master. So, here’s all the resources and options that is available for us to use (exploit and explore).
source: An Introduction to Deep Reinforcement Learning (huggingface.co)
Directly Available Resources
These are a must and the first and most important steps for building a foundation:
Model improvement resources in the console (https://aws.amazon.com/deepracer/racing-tips/?nc=sn&loc=5)
Learning modules: These are two courses of 10 hours each and are very important for building fundamentals.(you must complete them)
Other articles on Medium, but most of them don’t focus on the student league.
Some great YouTube channels:
BoltronRacingTeam: https://www.youtube.com/@BoltronRacingTeam/
AWSDeepRacerCommunity:https://www.youtube.com/@AWSDeepRacerCommunity
Joining the community
This is the most important part of the whole journey. Joining Discord will give you a lot of insights, and there are a lot of people to help you throughout your journey. Asking questions and learning from others and sharing insights of what you have learnt will help you very much in this journey.
join by clicking here -> https://discord.com/invite/G72rNQmJRg
Gaining Insights from:
Training hours
Every person gets 10 hours, and you have to use it carefully. The ideal training time depends on the reward function.
The more simple the reward function, the less time it takes to train. You must train at least for 2 hours and can go up to 10 hours.
Pro tip: Cloning a model doesn’t help much. Instead train once for a long time. sometimes cloning a long trained model for an hour or half also helps
Bonus: You can also train for unlimited hours for free if you have a good configuration PC. By using deep-racer-for-cloud, there are various resources on the web that show how to set this up. This is only useful for testing purposes, as in the end its compulsory to train your model in the Deepracer student console. https://aws-deepracer-community.github.io/deepracer-for-cloud/
Action space and speed
The action space used in the student league limits the speed to [0.5–1] m/s. Speed is a very important factor, we must maintain it to 1m/s.
Track insights
The length of the track, the number of turns, and the waypoints and steps provided not only help to calculate the number of steps, waypoints, and other useful information but also aid in developing a strategy to navigate the track.
Leaderboard
Watching the videos of top racers multiple times gives us essential insights, such as how they maintain speed, how their car turns, and how they run on straight and curved sections of the track.
Downloading models
The CSV files help students to compare different models trained by them and provide a few insights, such as the reward received per step, the number of steps taken, etc.
Conclusion
AWS DeepRacer is a great tool for students who are looking to learn about machine learning through practical, hands-on experience. By participating in the Student League, students can gain valuable insights and develop their skills in reinforcement learning. To achieve success, students must utilize all available resources,
For more information visit:
In the next article, we will discuss how to build and analyze the Reward Functions
https://anshtanwar.hashnode.dev/craft-a-powerful-reward-function-in-student-league-81925c56a11e