kazeruch / super-mario-bros-v0

An Reinforcement Learning agent designed to learn and complete OpenAI Gym Super Mario Bros environment. These environments allow 3 attempts (lives) to make it through the 32 stages in the game. The environments only send reward-able game-play frames to agents.

Super-Mario-Bros-v0

A Reinforcement Learning agent designed to learn and complete the OpenAI Gym Super Mario Bros environment. These environments allow 3 attempts (lives) to make it through the 32 stages in the game. The environments only send reward-able game-play frames to agents.

OpenAI Gym

OpenAI Gym is a toolkit for developing and comparing reinforcement learning algorithms. It supports teaching agents everything from walking to playing games like pong or pinball. Gym is an open source interface to reinforcement learning tasks.

Reinforcement learning algorithms

• Double Deep Q-learning (off-policy, model-free)

Demo video

https://www.youtube.com/watch?v=O2QaSh4tNVw

Requirements

• Python 3.6 or 3.7
• CUDA Toolkit 10.1
• cuDNN v7.6.5
• Pipenv

How to install the packages

You can install the required Python packages using the following command:

• pipenv sync

How to run it

You can run the script using the following command:

• pipenv run python super_mario_bros_v0_ddqn.py

Note for developers

• you are responsible to update the default values of the hyperparameters
• research will continue once I get better hardware