Learning Rules for Cooperative Solving of Spatio-Temporal Problems
Daan Apeldoorn · 2015
This paper addresses the issue of creating agents that are able to learn rules for cooperative problem solving behavior in different multi-agent scenarios. The proposed agents start with given rule frag- ments that are combined to rules determining their action selection. The agents learn how to apply these rules adequately by collecting rewards retrieved from the simulation environment of the scenarios. To evaluate the approach, the resulting agents are applied in two different example scenarios.