Dynamic Continuous Distributed Constraint Optimization Problems

Abstract

The Distributed Constraint Optimization Problem (DCOP) formulation is a powerful tool to model multi-agent coordination problems that are distributed by nature. While DCOPs assume that variables are discrete and the environment does not change over time, agents often interact in a more dynamic and complex environment. To address these limiting assumptions, researchers have proposed Dynamic DCOPs (D-DCOPs) to model how DCOPs dynamically change over time and Continuous DCOPs (C-DCOPs) to model DCOPs with continuous variables and constraints in functional form. However, these models address each limiting assumption of DCOPs in isolation, and it remains a challenge to model problems that both have continuous variables and are in dynamic environment. Therefore, in this paper, we propose Dynamic Continuous DCOPs (DC-DCOPs), a novel formulation that models both dynamic nature of the environment and continuous nature of the variables, which are inherent in many multi-agent problems. In addition, we introduce several greedy algorithms to solve DC-DCOPs and discuss their theoretical properties. Finally, we empirically evaluate the algorithms in random networks and in distributed sensor network application.