An explicit semantics for coordination multiagent plan execution
Jaeho Lee · 1997
An agent coordinating with another needs a model of the other agent to avoid interference or to promote cooperation. Generally, a model consists of both declarative information such as goals and beliefs, and procedural knowledge such as agents' plans. Agent plans in their executable form, however, are not well suited for reasoning over them, particularly because the plans are meaningful only when they are interpreted with the execution or operational semantics. We have developed Structured Circuit Semantics (SCS) and Grafcet Model of Agent Plans (GAP) to distill out explicit, transferable operational semantics for agent plans from embedded implicit semantics of plan specifications used by plan execution systems such as our University of Michigan Procedural Reasoning System (UM-PRS). While SCS makes explicit the directives to an interpreter, GAP makes explicit what behavior these directives elicit. Our explicit, transferable semantics allows us to define what situations and transitions occur in plan executions, and to capture the dynamically evolving situation transitions in the course of multiagent plan execution. In turn, our classification of situations enable coordinating agents to find commitment conditions to avoid interference situations, or to promote cooperation. In this thesis, we present and demonstrate this closed-loop process of exploiting our model of agent plans to discover multiagent coordination requirements, and then applying the coordination requirements back to the agent plans. First, we enumerate needed transferable semantics and present a language for representing the identified semantics. We then introduce a process model for defining semantic implications in an interpreter and present an algorithm to use the process model for multiagent coordination.