Before I will do the checklist I wanted to ask for feedback, do the community think that this change is useful and want me to continue?
Added BabyAI-Text environments to Minigrid, which provide textual descriptions of observations instead of symbolic representations. This extends the BabyAI platform to enable training language models as agent policies in interactive environments. The new environments maintain the same underlying mechanics as original BabyAI tasks but return text observations describing the agent's partial view of the environment.
This addition enables research on grounding language models in interactive environments while maintaining the sample efficiency.
Type of change
Please delete options that are not relevant.
[x] New feature (non-breaking change which adds functionality)
[x] This change requires a documentation update
Screenshots
Example observation:
Corresponding description:
[
'You see a wall 2 steps right',
'You see a blue key 3 steps left and 1 step forward',
'You see a red box 3 steps left',
'You see a blue box 2 steps left and 1 step forward',
'You see a red ball 1 step left and 1 step forward',
'You see a grey ball 1 step forward',
'You see a blue box 1 step right and 1 step forward'
]
Checklist:
[ ] I have run the pre-commit checks with pre-commit run --all-files
[ ] I have commented my code, particularly in hard-to-understand areas
[ ] I have made corresponding changes to the documentation
[ ] My changes generate no new warnings
[ ] I have added tests that prove my feature works
[ ] New and existing unit tests pass locally with my changes
Description
Before I will do the checklist I wanted to ask for feedback, do the community think that this change is useful and want me to continue?
Added BabyAI-Text environments to Minigrid, which provide textual descriptions of observations instead of symbolic representations. This extends the BabyAI platform to enable training language models as agent policies in interactive environments. The new environments maintain the same underlying mechanics as original BabyAI tasks but return text observations describing the agent's partial view of the environment.
Code is copied and adapted from https://github.com/flowersteam/Grounding_LLMs_with_online_RL
Key additions:
This addition enables research on grounding language models in interactive environments while maintaining the sample efficiency.
Type of change
Please delete options that are not relevant.
Screenshots
Example observation:
Corresponding description:
Checklist:
pre-commitchecks withpre-commit run --all-files