ms-pacman-agent

Project 3 - AI course (ECSE 526) - McGill - Fall 2016

Check project instuctions here.

In this project, I coded a reinforcement learning agent whose role is to play the famous Ms. Pacman game within the Atari emulator (Stella) using the Arcade Learning Environment ALE.

The agent coded for this project uses RL-Glue interface to communicate with the ALE. The C++ code could be found in the pacman_rlglue folder. The complete results of the simulation could be found in the results_30_episodes file. The agent was able to achieve a mean score of 4688 over the 30 episodes with a maximum of 8980. Please check the assignement report for details about performance and the reward funcion used.

Installation instructions

Install RL-Glue core from here
Install RL-Glue C/C++ codec from here

1. Compilation of Environment (ALE)
   RL Glue core and C/C++ codec must be installed on the system (links given above)

   sudo apt-get install cmake libsdl1.2-dev
   cd <project_root_folder>
   cmake -DUSE_SDL=ON -DUSE_RLGLUE=ON -DBUILD_EXAMPLES=ON
   # ALE needs to be compiled with -DUSE_RLGLUE = ON to use rlglue controller
   make -j 4

   For further details, check the Manual for ALE

2. Compiling the Agent and the Experiment
   Take the example of Ms. Pacman

   cd <project_root_folder>
   cd pacman_rlglue
   make

Start the game

Now we need to start 4 processes from the Project root folder as follows:

rl_glue
./pacman_rlglue/pacman_agent
./pacman_rlglue/pacman_experiment
./ale -display_screen true -game_controller rlglue roms/ms_pacman.bin