Kodi Community Forum
[GSoc 2020] Reinforcement learning bots for RetroPlayer (Draft) - Printable Version

+- Kodi Community Forum (https://forum.kodi.tv)
+-- Forum: Development (https://forum.kodi.tv/forumdisplay.php?fid=32)
+--- Forum: Kodi Application (https://forum.kodi.tv/forumdisplay.php?fid=93)
+---- Forum: GSoC (https://forum.kodi.tv/forumdisplay.php?fid=299)
+----- Forum: GSOC 2020 (https://forum.kodi.tv/forumdisplay.php?fid=301)
+----- Thread: [GSoc 2020] Reinforcement learning bots for RetroPlayer (Draft) (/showthread.php?tid=352554)



[GSoc 2020] Reinforcement learning bots for RetroPlayer (Draft) - AlexBogdan - 2020-03-15

[GSoc 2020] Reinforcement learning bots for RetroPlayer (Draft)

This is my draft version, so I'll be editing this post with more information when I'll have a clear idea about the project.

About me
 
   Hello! My name is Alexandru Bogdan Andrei, and I'm pursuing an M.S. (first-year) in Artificial Intelligence at the University "Politehnica" of Bucharest. Searcing for a GSoC project, I've found this really interesting one here on Kodi's project ideas list, and I would like to discuss a little bit about it since I feel that it is on my profile. This project could represent for me the opportunity to implement a real project with everything that I've learned during my first CS years.
   During my B.Sc final year, during the AI and ML courses, I've had 2 homework related to this topic. Also, my Bachelor's Thesis had a bot that learned how to play a 2D game (surviv.io) using Computer Vision and Machine Learning. I'll probably present more about them in my official proposal if it's required. This semester I'm studying Multi-Agents Systems, a course that again could be helpful in this project.

Contact
Name: Alexandru-Bogdan Andrei
E-mail: [email protected]
Facebook: alexbogdan.andrei

Summary
A somewhat small but explanatory walkthrough of the project. It should not be overly detailed just enough to understand the problem trying to be fixed and how this project opt to solve it.

How will I achieve this
Explain how the project will be done, what technologies are needed and how to implement them.

What will the project focus on
Explain what the project will focus on, what is the important parts of the project.

Benefits
Who will benefit and why from this project? Think about what a user or developer may need or do to benefit from it. Why does it benefit many users?

Goals
What is the goal of the project, a project may not always solve the problem entirely as it may take to much time. Think hard about what can be accomplished during a summer with your skill and deduct that quite a bit. If the project can't be done after this perhaps its better to opt for a smaller one or one with subgoals.

What does it touch in Kodi
Think about what parts of the code this may touch, Kodi is a big application and a lot of the code is spread out and very complex. If a project touches to much it may be impossible to complete.

Requirements
What is needed to complete the project, what code language knowledge what hardware etc.

Possible mentors
garbear?, velocity?


RE: [GSoc 2020] Reinforcement learning bots for RetroPlayer (Draft) - AlexBogdan - 2020-03-15

I would like to know where should I start looking for the current game engine and if there's anything to be done right now (some bug fix or small stuff) to get in touch easier with the code.


RE: [GSoc 2020] Reinforcement learning bots for RetroPlayer (Draft) - Prof Yaffle - 2020-03-15

Ping @velocity @garbear


RE: [GSoc 2020] Reinforcement learning bots for RetroPlayer (Draft) - Klojum - 2020-03-16

@AlexBogdan

Just a thought... Using Comic Sans these days will not look good on your curriculum vitae Wink


RE: [GSoc 2020] Reinforcement learning bots for RetroPlayer (Draft) - AlexBogdan - 2020-03-16

(2020-03-16, 00:13)Klojum Wrote: @AlexBogdan

Just a thought... Using Comic Sans these days will not look good on your curriculum vitae Wink

Yeah, I thought that I can edit the post after submission, but I didn't found any option... I'll definitely use a simpler one on my second try Smile)


RE: [GSoc 2020] Reinforcement learning bots for RetroPlayer (Draft) - Hitcher - 2020-03-16

Fixed the font for you.


RE: [GSoc 2020] Reinforcement learning bots for RetroPlayer (Draft) - garbear - 2020-03-18

Hi Alexandru,

Cool to see you're studying AI. Are you familiar with OpenAI retro gym? that's basically what this idea was about. How familiar are you with reinforcement learning? Did you read the Reinforcement Learning Bots github issue? What do you think of it? Do you think you can start with the Q-function and derive the emulation equations? I plan to scan upload my work to derive the emulation equation, but I might need help re-interpreting it after three years Smile . I know these are open ended questions, just figured I'd start narrowing down topics for your draft.

Regards,
Garrett


RE: [GSoc 2020] Reinforcement learning bots for RetroPlayer (Draft) - AlexBogdan - 2020-03-22

Hello and sorry for a late response.

I do have some experience with the OpenAI gym. Also, about Reinforcement Learning, I do have only my academic experience, nothing from real projects. About the Q-learning part, I did had homework that had a simplified idea similar to this project. I'll try to come as fast as I can with a proposal for this project, so I could get some feedback.


RE: [GSoc 2020] Reinforcement learning bots for RetroPlayer (Draft) - garbear - 2020-03-24

Do you plan to develop your project proposal above?