Customize BabyAI-Text

[GoooKuuu]

Hello, thank you so much for sharing this amazing work. I am using it for my project. However, I've observed that the current version is limited to supporting only a specific set of environments. Considering the potential and flexibility of BabyAI-Text, I am keen to extend its application to a wider range of environments, such as Key Corridor.
I would greatly appreciate any guidance or advice on how to effectively customize BabyAI-Text for additional environments. If there are any available documentation, tips, or starting points that could aid in this endeavor, it would be immensely helpful.
Thank you once again for your dedication to this project and for providing such a valuable resource to the community. I eagerly anticipate your guidance and am excited about exploring further possibilities with BabyAI-Text.

[ClementRomac]

Hi,

It would be great to see BabyAI-Text extended to other tasks!

As it has been initially made for our experiments on grounding, the current code mapping the grid environment to text is only working for single-rooms setups and is quite muddy (I apologize for that).

What this piece of code does is simply parsing the observation (a 7x7x3 array) and generating sentences given objects in it. It only works in single-room environments as it does not look for objects the agent can see through an opened door. I guess this is the first major improvement you could make. Maybe @tcarta can give other inputs on this.