Source code for the IJCAI-23 paper Multi-Agent Intention Recognition and Progression, by Michael Dann, Yuan Yao, Natasha Alechina, Brian Logan, Felipe Meneguzzi and John Thangarajah.
To install the Python requirements via Anaconda, use
conda env create -f environment.yml
.
Feel free to email Michael at [email protected] if you have any issues getting the code to run.
To recreate the results from the paper, run:
python python_agent.py scenario_name
where scenario_name is one of {neutral_1, neutral_2, neutral_3, neutral_4, allied_1, allied_2, allied_3, adversarial_1, adversarial_2, adversarial_3}.
Agent scores are automatically logged to results/cooperative_craft_world_dqn/results_scenario_name.csv.
To train a new RL policy, run:
python python_agent.py train
The goal items for the RL policy can be configured in scenario.py (line 32).