AASERT93/Real-Time Control of Reasoning.

Abstract : This AASERT award supported students working on the development of next-generation planning and scheduling systems. Students supported by the award contributed directly to both the development and implementation of these systems. The dissertations completed involved the theoretical investigation of reason maintenance and dynamic backtracking and their practical impact on search engines, and the development of a new search technique known as 'limited discrepancy search' that has been successfully implemented at a variety of external sites. Limited discrepancy search has also been incotporated in a CIRL-developed tool that develops manufacturing schedules, and the technique has produced the best known solution on realistic problems related to aircraft manufacture. Another AASERT-supported student interfaced the scheduling tool to Microsoft Project, demonstrating both the flexibility of the method and its applicability to a wide range of problems. The parent AFOSR award involved planning research as well, and two of the five supported students have worked on problems fundamental to the development of more effective automated planning systems. This has included the development of specialized planning tools for simplified domains and an investigation of the reasons that these systems are effective, and a formal investigation of the role of causality in representations of commonsense knowledge about actions, and the impact such representations will have on planning systems.