Multi-agent multi-goal problems where multiple goals are assigned to multiple agents in an environment have applications in various domains. In this paper, we explore an assignment problem where greedy agents looking to maximize their reward in the environment are aware of the goals and their properties, but not of other agents. We aim to formulate a reward structure that encourages the greedy agents to behave in a manner similar to a central planner making the optimal assignments for them. The central planner is making assign- ments to minimize the social cost metrics such as the total distance travelled by all the agents, or the total waiting time for all the goals. We model and analyse a reward structure that makes the greedy agent pick assignments that resemble the central planner’s allocations, and compare its social cost and performance with that of the optimal assignment.
To run the simulation, execute python game.py