Plan recognition: From single-agent to multi-agent plans

Abstract

The ability of recognizing a multi-agent plan — a plan which has been generated for multiple executing agents — implies the skill of recognizing the plan underlying a collective activity in which a group of agents is involved. Multi-agent plan recognition is a necessary and useful extension of current PR systems in some domains.

In this work starting from CHAPLIN (CHart Applied to PLan INference, a chart-based plan recognition system) we discuss how extending this system to the multi-agent world, many problems arise. More in general we consider the challenges that the multi-agent domains produce for the plan recognition systems. In fact this extension involves non-trivial problems and it calls for a general theory of the relationships between single-agent plans and multi-agent plans.

CHAPLIN shows how the chart allows an effective representation of the structural organization of plans.

Some of the main problems arising in MAP recognition are identified and discussed (iterative decomposition, virtual actions, virtual plans). We show how Multi-CHAPLIN -the extension of the CHAPLIN system — can partially manage those problems.

Finally, we will call for the necessity of a theory of “plan transformation”: from single-agent plans to multi-agent plans. Such a theory is claimed to be very advantageous both for the generality and abstraction of plan formalization, and for the economy of plan memory.