profintegra/MultiAgentBattleSimulation

multi agent reinforcement learning environment

MultiAgentArmySim

Multi agent reinforcement learning environment where teams of agents, each with a neural network, work together to defeat the other teams in an all out battle.

Installation:

python setup.py install

How to use:

usage: run_server.py [-h] [--grid_size GRID_SIZE] [--num_agents NUM_AGENTS]
                 [--num_teams NUM_TEAMS] [--num_episodes NUM_EPISODES]
                 [--episode_length EPISODE_LENGTH]
                 [--saved_model_path SAVED_MODEL_PATH] [--train_model]
                 port

Multi agent battle simulation

positional arguments:
	port                  enter a valid port for the server to listen to

optional arguments:
	-h, --help            show this help message and exit
	--grid_size GRID_SIZE 
			grid size (must be an integer)
	--num_agents NUM_AGENTS
                    	number of agents in each team
	--num_teams NUM_TEAMS
                    	number of teams
	--num_episodes NUM_EPISODES
            	        number of episodes
	--episode_length EPISODE_LENGTH
            	        length of each episode
	--saved_model_path SAVED_MODEL_PATH
            	        path to pretrained neural network
	--train_model         boolean
			train model

# Example with grid size 100, 200 agents on each team, 1 episode with length 100, and a pretrained neural network model. Agents use REINFORCE to train the model
python run_server.py 8888 --grid_size 100 --num_agents 200 --num_teams 2 --num_episodes 1 --episode_length 100 --saved_model_path src/saved_models/test_model.pth
# then go to localhost:8888 to see the agents duke it out!