abstract
The aim of the project is to implement Pommerman, a modified version of the classic game Bomberman by Nintendo (1983).
The game consists in four agents that can either cooperate in teams of two in the Co-op mode or can be opponents in the free for all (FFA) mode. The goal of the agents is to place bombs on the board and try to eliminate the opponents so that it remains at least one member of either team alive or only agent alive in the FFA version.
To implement this project, we are going to use Jason, a suitable framework for developing multi- agent systems. Then, for the learning part of the project, since the Reinforcement Learning out-of-the-box algorithms are shown to be not satisfactory [REH+22] for this particular game, our purpose is to find an algorithm that could get us better results.
outcomes