jiseonghan / super-mario-rl

🍄Reinforcement Learning: Super Mario Bros with dueling dqn🍄

🍄 Super-Mario-RL

This is a private project to make Super Mario Agent.

It consists of training an agent to clear Super Mario Bros with deep reinforcement learning methods.

Here are my super mario agents with dueling network. ( trained 7,000 epoch )

Get started

Cloning git

git clone https://github.com/jiseongHAN/Super-Mario-RL.git
cd Super-Mario-RL

Install Requirements

pip install -r requirements.txt

Or Install Manually

• Install openAI gym

pip install 'gym'

• Install Pytorch

pip install torch torchvision

• Install nes-py

pip install nes-py

• Install gym-super-mario-bros

pip install gym-super-mario-bros

Running

Train

• Train with dueling dqn.

python duel_dqn.py

Result

• score.p : save total score every 50 episode
• *.pth : save weight of q, q_target every 50 training

Evaluate

• (Now, pre-trained agent has been corrupted😢)
• Test and render trained agent.
• To test our agent, we need 'q_target.pth' that generated at the training step.

python eval.py

• Or you can use your own agent.

python eval.py your_own_agent.pth

Reference

Wang, Ziyu, et al. "Dueling network architectures for deep reinforcement learning." International conference on machine learning. PMLR, 2016.

super-mario-rl's Issues

Inquiry About Training Approach for Super Mario Agent

Hi,

I recently came across your fascinating project on GitHub where you've developed an agent to play Super Mario using reinforcement learning. Your work is truly impressive, and I'm very interested in understanding more about your training methodology.

I was wondering if you could share some insights regarding how you trained your agent:

Did you train the agent on individual levels separately, or was it trained across the entire game directly?
Is the trained model capable of completing the entire Super Mario game without modifications, or does it require level-specific adjustments?
Your answers would greatly help in enhancing my understanding of applying reinforcement learning in complex environments like Super Mario. Thank you very much for your time and for sharing your work with the community.

Best regards,
asyuaye

How can I get a video like mario1.gif?

Hi, Thanks for wonderful project!

I want to study RL via this code, but I have a question.

How can I get a video like mario1.gif? because I want to know my agent is trained well via gif or video.

thanks.

Untrained Model in the last commit.

It seems, q_target.pth model is overridden by my model which you merged with the pull request. Actually i just tried for 5 episodes. Get back 39th commit model (your model).

And the q_target.pth in 39th commit having EOF file error. This Model size is 2MB (approx) but it should be 41 MB (approx). So there is mismatch is the state_dict() size.