Stochastic recruitment: Controlling state distribution among swarms of hybrid agents

摘要

This paper introduces a control architecture for centrally controlling the ensemble behavior of many identical agents. A swarm of robots or other agents performing a variety of tasks is often modeled as a collection of hybrid-state agents, whose discrete switching behaviors are controlled by finite state machines. The number of agents in the swarm in a particular discrete state is a function of the rate at which agents transition between state. These state transitions are often modeled as stochastic interactions with the environment. We show that effective control over the distribution of agents in each discrete state can be achieved by designing the agents to transition between tasks randomly, according to a centrally determined state transition probability graph. The centrally-determined policy varies with time and with feedback information by rebroadcasting the probability graph to all agents. Feedback policies will be presented for the case in which the central controller has limited or no knowledge of the states of each agent.