Atari2600 Academic research (reinforcement learning)

[spaceInvader89]

Hi!

I'm a CS masters student currently working on my thesis in reinforcement learning.

There's a really cool 2600 game available here called Skeleton+ by Eric Ball (released in 2002).

Because of it's game properties Skeleton+ is perfect for my AI research (This research, if successful, might wind up as a small paper in some AI magazine or blog).

Legally I can't publish anything using this game without permission from Eric Ball (or from AtariAge?) - but I didn't manage to reach neither via mail.

Can anyone help me to reach either Eric or an admin from AtariAge regarding this issue?

 

Thanks :)

Nir Weingarten

[Big Player]

Are you sure you need permission? I'm not a lawyer or know much about the academic world.  Usually homebrew developers don't want someone else selling or distributing their work without their permission. I haven't heard of any not wanting their games used in academic research.

 

These annotations are new to me, but lets try them out.

 

@EricBall   @Albert

[spaceInvader89]

Hi @Big Player

Thanks for helping.

Also not sure if I need or don't, but seems like the right way to go. Also it would be great having the developers see the AI playing their games :)

[EricBall]

@spaceInvader89

Which email address did you use for me?  (The old atariage one stopped working years ago.)

 

But I have no issues with you using the game itself as part of your research.  (That would fall under fair use anyway, but I appreciate you asking first.)

 

Send me a DM if there's anything in particular you'd like to know about the game which would help you out.  I'm also curious about what particular properties Skeleton+ has which caught your interest.

 

Just don't use it as AI training for killer robots.

[spaceInvader89]

@EricBall thanks.

Yes, iv'e used the old AtariAge email..

Apart from being a balanced game and lot of fun Skeleton+ also uses stereo audio.

The research will incorporate audio in two channels into a neural net and we'll measure the differences in performance while using no audio, mono audio and stereo audio on various RL algorithms.

I'll DM you once I have some progress done, should be interesting!