Binary max-sum for clustering-based task allocation in the RMASBench platform
Abel Correa · 2016
Multiagent systems has been extensively used in the context of urban disaster situations. In such situations, the agents have to perform tasks in order to mitigate the damage in a simulated city, dealing with uncertainty and conflicting information during the disaster management. This paper addresses the group formation in the RMASBench platform centred on clustering-based task allocation. The RMASBench provides an API for multiagent coordination based in the distributed constraint optimization problem. We present an objective function to minimize a distance metric based in the similarities between the features of the agents and the set of perceived tasks. To efficiently optimize our objective function, we use a message passing algorithm in a graphical model representation. Our approach is compared with the state-of-art approaches in the domain of the RMASBench platform and our results show that is possible to reduce the amount of exchanged messages and constraint checks, improving the performance of the agents in some situations.