Add ls20 env - #625
Open
mahiuchun wants to merge 2 commits into
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A port of https://arcprize.org/tasks/ls20 probably the most well known public game in ARC-AGI-3
I did it using codex and gpt-5.6-sol asking it to translate the python version of the game; I think the code quality is okay.
Animation frames that can't be acted upon are not included right now.
I added some quality of life improvements:
--env.fpsas I find the default 30 FPS (from Python version) a bit too fast--env.reset-enabled(default off) as the RESET action can be confusing for the RL algorithm or make it look like the random agent never loses a life (since it would hit RESET all the time)--demoin the main function showing a winning sequence. Otherwise the main function is manual playI tried training as well.
So far the best I can get is clearing the very first level.
This game is essentially a maze with twists so it might provide some research value as well.
It is not the hardest game in ARC-AGI-3 for RL algorithms as the action space is small (4 arrow keys + optional RESET); games where the action space is clicking anywhere in the 64x64 canvas is even harder