Multiagent planning as control synthesis

This paper proposes a new multiagent planning approach to coordination synthesis that views distributed agents as discrete-event processes. The connection between discrete-event control synthesis and coordination planning is first established, thereby enabling the exploitation of the vast body of knowledge and associated software synthesis tools from 'The Supervisory Control of Discrete-Event Systems' for automatic coordination synthesis of distributed agents. Importantly, these coordinating agents designed collectively generate a behaviour guaranteed not to contradict any specified inter-agent constraint, is nonblocking and optimal. A simple planning methodology is proposed in terms of procedures supported by CTCT, an existing, freely available design tool developed based on the control synthesis framework. A simple example illustrates the use of the CTCT-based methodology to synthesize coordination modules for distributed agents. Discussions in relation to previous work examine the relative significance of the new multiagent planning framework.