Skip to content

This is an effecient implementation of the OpenAI gym's cartpole AI using a NeuroEvolution Algorithm (Similar to NEAT) which outperforms the traditional RL Techniques.

Notifications You must be signed in to change notification settings

SurenderHarsha/OpenAIGym-Cartpole-Evolution-

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 

Repository files navigation

OpenAIGym-Cartpole-Evolution-

This is an effecient implementation of the OpenAI gym's cartpole AI using a NeuroEvolution Algorithm (Similar to NEAT) which outperforms the traditional RL Techniques.

This is based on https://blog.openai.com/evolution-strategies/ But the cartpole environment is too simple to find significant difference between ES and RL techniques. Therefore the maximum reward and steps have been increased to 15000. The reason for this is that Evolution strategies may get lucky and converge to a solution in just 2 episodes as tested. Hence I have increased the time of gameplay.

On performing tests, the Algorithm takes approximately on an average of 75 episodes to converge to a solution of 15000.

It is based on the evolution of weights of a perceptron using genetic algorithms. For now to Solve for a score of 195, it takes about 1-40 Episodes randomly.

Just run the program until it says done. it then prints DONE, the weights used to calculate the action, the maximum reward and the number of episodes it took.

About

This is an effecient implementation of the OpenAI gym's cartpole AI using a NeuroEvolution Algorithm (Similar to NEAT) which outperforms the traditional RL Techniques.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages