Multi Agents Area Exploration

Quick Summary

This project is investigating how to explore an unknown building that its boundaries are known in prior. To analyze the efficiency of various exploration strategies, we generated sixteen different exploration method that is being used by the system homogenously as well as different starting and goal conditions. It is assumed that a reliable communication and map localization occurs in the background without considerable errors.

Furthermore, we also examined how agent loss would affect the overall performance by placing mines over the ground.

Installation of the Environment:

git clone https://github.com/DIRECTLab/Multi_Agents_Area_Exploration.git
cd Multi_Agents_Area_Exploration/
conda create -y --name multiagent python==3.9.2
conda activate multiagent
pip install \
	numpy==1.23.3 \
	opencv-python==4.6.0.66 \
    opencv-contrib-python==4.6.0.66 \
	pygame==2.3.0 \
	scipy==1.10.1 \
	jupyter_client==7.0.6 \
    jupyter_core==5.3.0 \
    numba==0.56.4 \
    pillow==9.2.0 \
    scikit-learn==1.2.2 \
    matplotlib==3.5.3 \
    pandas==1.4.4 \
    tqdm==4.64.1 \
    psutil==5.9.4

Running the Program:

To run the simulation, run the following command:

python main.py

If you would like to change the simulation parameters, you can do so by changing the values in the parameters_cfg.py file.

Simulation Parameters:

Parameter	Description	Default Value
Debug	Debug mode, renders a GUI	False
Create_gif	Create gif, Debug must be True to run this	False
agent_count_list	List of agent counts to run the simulation with	[2,4,6] or list(range(4,12,4))
iteration_repeat_experiment	Number of times to repeat the experiment	1 or list(range(0, 60))
min_rom_size	Minimum size of the room	30 or list( range(4,10,2))
Method_list	A list of all the Methods that the simulation will run, these all inherit from the base Agent class	-
Start_scenario_list	starting locations for all the agents	Manual_Start, Rand_Start, Edge_Start, Top_Left_Start, Center_Start, Distributed_Start
Start_Goal_list	initial goal locations	Manual_Start, Rand_Start, Edge_Start, Top_Left_Start, Center_Start, Distributed_Start,
Robot_Loss	Holds a list of classes that change the scenario, where Agent is a normal run, Unrecoverable the robots will hit random mines and become disabled, and last the Disrepair robots can help and fix other robots	[Agent, Unrecoverable, Disrepair]

Method	Description
Frontier_Random	Each agent randomly selects a frontier section of the map to explore.
Frontier_Closest	Each agent selects the closest frontier section of the map to explore.
Unknown_Random	Each agent randomly selects an unknown section of the map to explore.
Unknown_Closest	Each agent selects the closest unknown section of the map to explore.
Voronoi_Frontier_Random	The map is split up into various sections and then the frontier random method is used within each section.
Voronoi_Frontier_Closest	The map is split up into various sections and then the frontier closest method is used within each section.
Voronoi_Frontier_Help_Closest	The map is split up into various sections and then the frontier closest method is used in tandem with the paired searching strategy.
Voronoi_Frontier_Help_Random	The map is split up into various sections and then the frontier random method is used in tandem with the paired searching strategy.
Voronoi_Unknown_Help_Closest	The map is split up into various sections and then the unknown closest method is used in tandem with the paired searching strategy.
Voronoi_Unknown_Help_Random	The map is split up into various sections and then the unknown random method is used in tandem with the paired searching strategy.
Decision_Frontier_Closest	Each agent shares it's goal position in the mutual data set. As each agent sets a new goal position it checks to see if any other agent is already visiting that location. If not, it keeps that location. If it is already taken |then the agent with the shortest path keeps that location and the other is assigned a new goal.
Decay_Epsilon_Greedy_Unknown	Each agent is assigned to either random or closest exploration of a frontier based on an epsilon value. As the simulation progresses the epsilon value decays resulting in less exploration as the map is explored more.
Decay_Epsilon_Greedy_Frontier	Each agent is assigned to either random or closest exploration of an unknown area based on an epsilon value. As the simulation progresses the epsilon value decays resulting in less exploration as the map is explored more.
Epsilon_Greedy_Unknown	Each agent is assigned to either random or closest exploration of an unknown area based on a constant epsilon value.
Epsilon_Greedy_Frontier	Each agent is assigned to either random or closest exploration of a frontier based on a constant epsilon value.
AntiMajority	Allowing each agent to choose the anti-majority of search methods chosen by the other agents. The two options are random or closest.