Multiple Agent Autonomous Control A Hybrid Systems Architecture

Abstract

Hybrid systems are systems in which a digital control automaton receives sampled sense data about the state of a continuous plant and, occasionally, issues a change in the control law for a plant controller. The digital control automaton may fail to meet design requirements for local control of plant state trajectory, and need to be replaced on-line. This paper is an overview of our architecture for systems in which there is a plant (or system of plants called plant processes), each plant process has a plant controller with an Agent for its digital control program and plant process controller. That Agent monitors its plant process and its plant process controller and its digital control program, passes and receives messages to and from other Agents for other plant processes, and occasionally computes and installs a new finite digital control automaton for its plant process controller. There is no supervisor. We call this Multiple Agent Hybrid Control. It is to be emphasized that the purpose of the Agent is to construct new finite control automata to replace old ones on the fly whenever performance and controllability and observability criteria are violated, and that the Agents have no central supervisor. The tools used are distributed message passing from Agent to Agent of information about Lagrangians, and the Lagrangian variational-automata method for extracting the digital control automata of Kohn’s declarative control [20]. The underlying hybrid systems model is that of Kohn-Nerode [34].