To run the train.py script, navigate to the directory containing the script and run the following command with Config from reinforcement_learning/config.py:
python train.py
To track the progress of our experiments and view detailed metrics, visit our experiment tracking dashboard on Weights & Biases.
In the "results" folder, you can find the following images:
-
Visual Tile for Each Agent in NMMO Grid:
- Description: This image represents the visual tile for each agent in the Neural MMO grid, providing insights into the spatial representation of the environment.
-
Action Decoder of NMMO:
- Description: This is the action decoder architecture of the Neural MMO (NMMO) model.
-
Multihead Attention Communication Channel between Agents:
- Description: This is the communication channel between agents using multihead attention mechanism.
- The foundational work and inspiration for this project came from the nmmo baseline
- We drew multi head attention communication insights and ideas from the research paper Is Centralized Training with Decentralized Execution Framework Centralized Enough for MARL?