Automated Service Composition Using AI Planning and Beyond

Abstract

Automated Service Composition is one of the "grand challenges" in the area of Service-Oriented Computing. Mike Papazoglou was not only one of the first researchers who identified the importance of the problem, but was also one of the first proposers of formulating it as an AI planning problem. Unfortunately, classical planning algorithms were not sufficient and a number of extensions were needed, e.g., to support extended (rich) goal languages to capture the user intentions, to plan under uncertainty caused by the non-deterministic nature of services; issues that where formulated (and, partially addressed) by Mike, being one of his key contributions to the service community.In this chapter, we look at the development of the original vision of automated service composition as AI planning, going from planning with extended (rich) goals, further developing into composition under uncertainty, extending it to other domains (and reformulating the service composition as composition of sensors and actuators), and then showing possible alternative techniques for highly scalable service composition at the expense of the richness of the domain representation.