Our research aims to provide three core contributions to the area of automated planning: 1. A formal characterization of the conditions under which a complex plan remains viable, 2. A principled method of generalizing various plan representations, and 3. A unified policy representation that embodies a generalized plan and allows an agent to execute efficiently. We have demonstrated the potential of our approach for sequential and partial-order plans, and we are currently extending our work to a richer form of partial-order plan. In a dynamic environment, an intelligent agent must consider contingencies and plan for them. We aim to address this key issue by building more robust artificial agents through the generalization of a variety of plan forms.
展开▼