Abstract
We are interested in the development of planners that work in real-world and large-scale applications. Most of these applications require systems that perform and combine the usual planning tasks, for example, plan formation, plan execution, monitoring, and recovering from failure, in a problem-dependent way. Furthermore, in most cases, these systems must integrate I the various planning techniques proposed in the literature, such as planning ahead and deferred and reactive planning. In this paper we present a system (called MRG) that eases all these difficulties by representing explicitly all the planning tasks and combining them by using a small set of powerful control structures. At the moment, we are experimenting with MRG in a complex large-scale application that is briefly described in this paper.