In this final project, you will develop and train a reinforcement learning (RL) agent using the MAgent2 platform. The task is to solve a specified MAgent2 environment battle, and your trained agent will be evaluated on all following three types of opponents:
- Random Agents: Agents that take random actions in the environment.
- A Pretrained Agent: A pretrained agent provided in the repository.
- A Final Agent: A stronger pretrained agent, which will be released in the final week of the course before the deadline.
Your agent's performance should be evaluated based on reward and win rate against each of these models. You should control blue agents when evaluating.
See video folder for a demo of how each type of opponent behaves.
clone this repo and install with
pip install -r requirements.txt
See main.py for a starter code.
For further details on environment setup and agent interactions, please refer to the MAgent2 documentation.