Multi_Agent_AI

Multi-agent AI system to solve different Sokoban-like levels written in Python (client side).
Repository developed for the DTU course Artificial Intelligence and Multi Agent Systems.

Proposed levels

• SAKaren.lvl
  
• MAKaren.lvl
  

Run a level

The server-side environment, provided during the course, runs in Java. In a Unix shell:

java -jar "$SERVER/server.jar" -l "$SERVER/levels/$lvl" -c "python multi_sokoban/searchclient.py $method --max-memory $mem" -g 150 -t 300

where

• $SERVER is the path to this repository.
• $method is the search method (e.g. -astar).
• mem is the memory threshold to be used the program.

This command is exposed through a tiny script for convenience. For instance:

exe_serve.sh SAD1.lvl -astar

Objectives

• Run lvl1 without getting an error.
• Create required maps.
• Choose a theoretical framework -> PPDL | BDI | POP | HTL.
• Choose a method of communication -> online-planing, deadlocks avoidance.
• Find paper for Sokoban-like with the chosen framework and multiagent.
• Solve all the levels with the agent.
• Papers in AAAI style of 6 pages.
• Open the repository.
• Choose a license.

Installation

The python client was packed as a module to ease its use. Once cloned, it can be installed from source via pip.

git clone https://github.com/FjodBorg/Multi_Agent_AI.git
cd Multi_Agent_AI
pip install .

After that, it should have installed numpy and the package is now accessible as a regular python package.

import multi_sokoban

Uninstalling the package can be done via pip.

pip uninstall multi_sokoban

Code guidelines

Flake8 and black it.

pip install flake8 black