search

google-deepmind / acme

A library of reinforcement learning components and agents
Apache License 2.0
3.48k stars 424 forks source link

How to configure your own environment #167

Closed monkeycc closed 2 years ago

monkeycc commented 2 years ago

simulator： VirtuaNES game： Super Mario Bros WIN10-11

# VirtuaNES 
# https://www.emulator-zone.com/doc.php/nes/virtuanes.html

# Super Mario Bros ROM
# https://wowroms.com/en/roms/nintendo-entertainment-system/download-super-mario-bros./23755.html

Like playing Super Mario Bros with virtuanes

Tutorial required

Learning by demonstration

ACME How to configure the environment

1111111111111111
nikolamomchev commented 2 years ago

@monkeycc I don't fully understand what are you asking for from the description. Could you please elaborate?

monkeycc commented 2 years ago

For example, NES games, PC games, Android games

How do I configure the environment interface and interface with acme

Do not use gym environment

In their own games, their own game interface, reinforcement learning and simulation learning

nikolamomchev commented 2 years ago

Got it. Acme assumes your environment implements the dm_env interface. In order to try different games you'll have to implement a Python object which exposes the interface and interacts with the game. The best way to do that will depend on the game itself and what is your goal (e.g. do you have access to the interval state of the have or just pixels etc.).

This is, however, our of the scope of Acme so I'll close this issue.