Silvia Miksch, Andreas Seyfang
In dynamically changing environments a planning system does not have all the required information at the first place and the world state can change, rendering the original plan invalid. Consequently, taking planning to complex, real-world domains calls for close coupling of planning and plan execution, known as continual planning. In domains where a high degree of adaptability is crucial and little a priori knowledge is available, explorative approaches are suitable. In other domains where comprehensive knowledge is available and human life depends on plan execution, like in medical treatment planning, only approved and validated procedures are admitted. Even more, these procedures constitute a rich asset in the planning task. To combine the utilization of such knowledge with the flexibility necessary in real-world applications, the Asgaard system integrates time-oriented, skeletal planning with real-time monitoring. It features a monolithic framework for the creation, verification, execution, and critiquing of plans with extensive covering of temporal modeling and data abstraction. In this paper, we describe the monitoring and plan adaptation capabilities of Asgaard.
Keywords: Planning, Meta-Heuristics for AI
Citation: Silvia Miksch, Andreas Seyfang: Continual Planning with Time-Oriented, Skeletal Plans. In W.Horn (ed.): ECAI2000, Proceedings of the 14th European Conference on Artificial Intelligence, IOS Press, Amsterdam, 2000, pp.511-515.