This repository presents an implementation of CAREL (Cross-modal Auxiliary REinforcement Learning) on the BabyAI environment. BabyAI is a platform for studying and developing AI agents in a gridworld environment, and CAREL is a reinforcement learning framework that incorporates the concept of corss-modality to enhance agent learning.
In this repository, we have extended and modified the original BabyAI codebase to include the CAREL auxiliary loss. The integration of CAREL allows agents to explore their environment more effectively, ultimately improving their performance and sample effectiency.