Meta-level architecture for executing multi-agent scenarios

Scenarios, which constrain the behavior of agents, can be the interface between computer experts (agent system developers) and application designers (scenario writers), as well as the interface between scenario writers and agents. It raises a number of challenging issues to execute multi-agent scenarios: 1) How can scenario writers generate customized scenarios easily? 2) How can scenario writers monitor and control scenario execution so as to debug errors in scenarios? 3) How can agents negotiate for scenarios to achieve robust behavior against scenario errors? So, in this paper, we provide a web style CUI (Graphical User Interface) for scenario writers to customize scenarios. We propose a meta-level architecture for scenario writers to trace and control the execution of scenarios by observing scenarios, and for agents to negotiate with others as well as scenario writers for scenarios. The meta-level architecture is illustrated by an experimental multi-agent system of evacuation simulation.