https://github.com/DoguD/aws-deepracer

Reward function experiments for AWS DeepRacer.
https://github.com/DoguD/aws-deepracer

aws deepracer reinforcement-learning

Last synced: 2 days ago
JSON representation

Reward function experiments for AWS DeepRacer.

• Host: GitHub
• URL: https://github.com/DoguD/aws-deepracer
• Owner: DoguD
• License: mit
• Created: 2020-08-21T23:23:29.000Z (about 4 years ago)
• Default Branch: master
• Last Pushed: 2020-08-29T13:28:16.000Z (about 4 years ago)
• Last Synced: 2024-08-02T13:23:40.542Z (3 months ago)
• Topics: aws, deepracer, reinforcement-learning
• Language: Python
• Homepage:
• Size: 14.6 KB
• Stars: 4
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# AWS DeepRacer Reward Function Experimentation

These are my experimentation with [AWS DeepRacer](https://aws.amazon.com/deepracer/) training.

## Reward Functions

|Iteration|Name|Strategy| 

| :---: |:---:|:-----|

|1|["Complete"](./RewardFunctions/complete.py)|A "complete" model which considers distance_to_center, speed, and progress(exponentially) as rewards and is_offtrack as penalty.|

|2|["Centerline Following"](./RewardFunctions/line_speed_progress.py)|A version of "Complete" modal, which rewards based on distance_to_center and progress(exponentially) and decreases rewards if car goes too slow.|

### Inspirations

- [Scott Pletcher](https://github.com/scottpletcher) / [deepracer](https://github.com/scottpletcher/deepracer)

### Authors

- [Dogu Deniz Ugur(Full Stack Developer)](https://dogudenizugur.me)