OpenAI Retro Contest – Compilation of Reinforcement Learning Write-Ups

Compilation of all the Write-Ups which i found on Discord/Internet for the OpenAI Reinforcement Learning Challenge.

Leaderboard status after the contest has ended.

1. World Models applied to Sonic

Github link:

Write-Up based on the World Models paper:

2. Train a Reinforcement Learning agent to play custom levels of Sonic the Hedgehog with Transfer Learning

Write-Up on Transfer Learning. This implementation is for the 5th place in leaderboard with a score of 5522.

3. Using Deep Reinforcement Learning to Play Sonic the Hedgehog

View story at


Another algorithm based on the World Models paper:

4. world-models-sonic-pytorch

Implementation of World Models using pytorch.

5. OpenAI Retro – Collision Maps

Visual Approach.

6. Retro contest solution

Source code for #8 place in Leaderboard and score of 5365.36

7. My final submission: the Improved JERK.

View story at

Implementation based on JERK algorithm.

8. OpenAI Retro Contest – Guide on “How to Score 6k points on Leaderboard in AI Challange”

OpenAI Retro Contest – Guide on “How to Score 6k points on Leaderboard in AI Challange”

My try on the contest with tuning of the hyperparamters.

9. Agents for OpenAI Retro Contest

Source Code:


Blog Post:

Well documented approach using Rainbow DQN.


10. OpenAI Retro Contest Report

View story at

(15th place): Rainbow DQN with NoisyNet exploration and expert replay


Those are all the write-ups which i was able to find.

Thanks for sharing those write-ups!

Those are a great way to learn about Reinforcement Learning algorithms and how we could improve our approaches.


Leave a Comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.