Dynamic multi-agent coordination: RoboCops

Abstract

We develop a simulation-based test bed that researchers can use to compare the performance of algorithms for decentralized coordination of autonomous agents. The particular coordination problem that we consider is a pursuit-evasion game known as the robotic cop (or RoboCop) problem. In this game, a group of agents representing the police attempt to capture one or more fleeing suspects. Play is restricted to a (bounded) rectangular grid of cells. Both the cops and suspect are restricted to horizontal and vertical movement to adjacent cells, with the cops enjoying an advantage in movement rate. We assume that centralized control is not possible (so each agent must make decisions autonomously), that communication between agents is not possible, and that protocols for actions have not been established in advance by the agents. We develop some simple heuristics for determining agent movement, and then demonstrate ways in which the test bed can be used to analyze and compare these heuristics.