Plan-Authoring System based on Sketches, Advice and Templates

Dr. Karen L. Myers (PI)
Dr. Thomas D. Garvey
Dr. Peter Jarvis
Mr. Thomas J. Lee
Dr. Mabry Tyson
Dr. Michael J. Wolverton
Dr. Marie desJardins (alumnus)

Project Overview


The AI planning community has focused much of its effort during the past few decades on producing stand-alone technologies that replicate individual human planning capabilities. The community's primary objective throughout that period was to produce systems that would replace human planners. While several powerful automated AI planning systems have been developed, there has been limited success in transitioning them from the research arena to significant user communities in either the commercial or military sectors. This failure derives from two main problems: the systems are difficult to use, and they require complete, high-fidelity models of the domain for them to produce meaningful plans.

The aim of the Active Templates program is to develop user-centric planning tools that augment the skills of human planners rather than replace them. These tools will be designed to have ``zero entry cost'': they will enable effective planning without special skills or training on the part of users or the development of extensive knowledge bases. However, the tools will be incremental in that they will support incorporation of more technologically advanced capabilities that may require additional training or development of domain-specific knowledge bases. The level of effort invested in training or knowledge base development for advanced technical capabilities should be commensurate with the added value that they provide. Furthermore, users will retain control over their invocation and effects.

The PASSAT Plan Authoring System

To address the objectives of the Active Templates program, we propose to design, develop, demonstrate, and evaluate PASSAT, a user-centric plan authoring system grounded in the concepts of plan sketches, advice, and templates. PASSAT will enable users, working as individuals or in teams, to quickly develop plans that draw upon past experience encoded in templates, but that are customized to their individual preferences and the demands of the current situation. PASSAT will provide a spectrum of planning tools that vary in their technological sophistication, entry costs, and knowledge requirements.

At the heart of the proposed effort will be an interactive plan authoring capability in which users can directly construct and edit plans. Users will be able to draw upon a library of templates, to the extent they desire, to assist with the development of a plan. These templates will include standard operating procedures, planning process descriptions, and cases corresponding to actual (or generalized) plans developed for related tasks. Tools will be provided that enable users to retrieve templates that are relevant to their evolving plans and that provide critiques of plans relative to stored templates. A plan sketch facility will enable users to outline descriptions of plans that would then be filled out using templates designed for similar tasks. This process of sketch completion could be directed by the user, drawing upon templates to provide suggestions, or completely automated.

Plan authoring activities will be tracked and recorded to provide a rich process history. These histories will store the key operations performed along with relevant contextual information that may have influenced the design process. Certain forms of rationale for a plan will be inferred from the histories, which could be used to facilitate plan explanation, to track changes in the operating environment that could impact the viability of authored plans, and to guide the modification process in response to such changes.

The proposed system will provide a powerful yet flexible framework for developing plans. This flexibility is not without disadvantages, however. While automated planning systems guarantee that the plans they produce satisfy appropriate definitions of correctness, no such guarantees can be provided with a user-centric plan authoring system. For this reason, we propose to develop plan analysis tools that will aid users in detecting problematic aspects of their plans. These plan analysis tools will be grounded in model-based reasoning methods.

A process management component will enable multiple planners to work as a team by allocating and coordinating responsibility and resources for building plan components. We will develop the necessary information management and coordination techniques to enable such distributed application of PASSAT. Coordination will be achieved through the use of process plans that assign responsibility to planners with loose resource allocations and constraints provided as guides. Processes will be specified using the same basic plan authoring methods that lie at the core of PASSAT. A process controller will ensure that all processes are carried out in a timely fashion, and information is distributed among planners in accord with their knowledge requirements.

Expected Impact

We believe that PASSAT will greatly enhance the abilities of human planners. It will enable them to develop plans faster than is possible to do manually; furthermore, the plans will be better suited to their preferences and needs than plans produced by purely automated technologies, which are limited to the contents of their predefined knowledge bases. The rationale component of PASSAT will make it easier to modify plans at execution time as unexpected events occur and assumptions made at planning time are violated.

"A Mixed-initiative Framework for Robust Plan Sketching", K. L. Myers, P. A. Jarvis, W. M. Tyson, and M. J. Wolverton. In Proceedings of the 13thInternational Conferences on AI Planning and Scheduling , Trento, Italy, June, 2003.  

"Mixed-initiative Planning in PASSAT", K. L. Myers, P. A. Jarvis, W. M. Tyson, and M. J. Wolverton. In Proceedings of the 13th International Conferences on AI Planning and Scheduling  (Demonstration Track), Trento, Italy, June, 2003.  

"PASSAT: A User-centric Planning Framework", K. L. Myers, W. M. Tyson, M. J. Wolverton, P. A. Jarvis, T. J. Lee, and M. desJardins. In Proceedings of the the 3rd International NASA Workshop on Planning and Scheduling for Space, Houston, TX, October, 2002. 

"Toward a Theory of Qualitative Reasoning about Plans", K. L. Myers. In Proceedings of the ICAPS-04 Workshop Connecting Planning Theory with Practice, 2004.

"Prioritizing Planning Decisions in Real-World Plan Authoring", M. J. Wolverton. In Proceedings of the ICAPS-04 Workshop Connecting Planning Theory with Practice, 2004.

Development of the PASSAT system has been funded by DARPA's Active Templates program through contract F30602-00-C-0058 under the supervision of Air Force Research Laboratory - Rome.

Last modified: December, 2000