Skip to content

jgoriasilva/AWS_DeepRacer

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 

Repository files navigation

AWS DeepRacer Experience

Welcome to my AWS DeepRacer repository, where I share my experiments and what I learned during the classification phase for the CSBC 2024.

About AWS DeepRacer

AWS DeepRacer is a competition that challenges you to create a self-driving car using reinforcement learning. This competition is accessible through the AWS console, providing a hands-on experience to self-driving car technology and machine learning.

About Reinforcement Learning

Reinforcement learning, the core of AWS DeepRacer, allows the car to learn optimal driving strategies through trial and error. By receiving feedback from its actions, the car continuously improves its decision-making skills to navigate the track independently. Learn more about RL here.

What You'll Find in This Repository

In this repository, I share my insights and experiences from participating in the virtual AWS DeepRacer competition. This includes:

  • Hyperparameter Tuning: How to better adjust the parameters to optimize the performance of the model.
  • Reward Functions: How to design better reward functions that guide the car's learning process effectively.
  • Tips and Tricks: Practical advice and strategies that I found useful during my experimentation.

I hope this repository serves as a valuable resource for anyone interested in autonomous racing and reinforcement learning.

Experiments

In the folder experiments I have summarized some key insights from the most interesting experiments, which are presented below.

Iteration Model codename Idea
1 FollowMiddle Follow the middle and penalize for steering.
2 FM-FT-Continuous Fine-tune previous model using a continuous reward function.
3 FM-FT-Speed-Progress Fine-tune FM-FT-Continuous on speed and progress.
4 FM-FT-Speed-Progress-Steps Fine-tune FM-FT-Continuous on speed and ratio between progress and steps.
5 FM-FT-PS-Slower-Turn Fine-tune FM-FT-Speed-Progress-Steps by reducing its speed on sharp turns to increase stability.
6 FM-FT-PS-Slower-Turn-Faster Fine-tune FM-FT-PS-Slower-Turn by increasing its speed.
7 FM-FT-PS-Slower-Turn-Faster-2 Fine-tune FM-FT-PS-Slower-Turn-Faster by increasing its speed.
8 Faster-SteeringSpeedProgress Fien-tune FM-FT-PS-Slower-Turn-Faster by reducing its speed on 15 degrees and increasing its speed on 0 degrees. Removed reward for following the center, reduced punishment for being off track.

About

My experiments using AWS DeepRacer for CSBC 2024

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published