Continuous Planning and Execution

A research project funded by DARPA's JFACC program.

Principal Investigator
Dr. Karen L. Myers
AI Center
SRI International

 Additional Team Members:
Mr. Raymond de Lacaze
Dr. Peter Jarvis
Mr. Thomas J. Lee

Project Summary

SRI intends to design, build, demonstrate, and deliver a continuous planning and execution system that will provide the kind of plan generation and replanning capabilities required for situated agents in highly dynamic environments. Central to this system is the tenet that plans are dynamic, open-ended artifacts that must persist and evolve in response to an ever-changing environment. In particular, plans must be updated in response to new information and requirements in a timely fashion to ensure that they remain viable and relevant.

The proposed system, called the "Continuous Planning and Execution Framework" (CPEF), will provide several capabilities that support the view of plans as `living objects'. The planning process itself will be open-ended: rather than attempting to generate a complete solution to a problem, the planner will produce a partial, approximate sketch that will form the initial basis for further development. The limits of the sketch will be determined by reasoning about the situation in which the planning takes place, encompassing such factors as uncertainty about the current and projected states, the reliability of action models, user desires to emphasize or ignore certain parts of the problem, and considerations of timeliness.

Plan generation is only one component of the system. In addition, the CPEF will provide both planning-time and runtime adaptation of plans in response to changes in both the environment and user goals. The CPEF will include several novel plan revision technologies, ranging from correctness-preserving methods based on dependency-structure maintenance, to transformational methods that sacrifice guarantees of correctness for better performance characteristics. For runtime plan modifications, repair will be possible in a range of modes, from loosely to tightly coupled interactions with the execution component. Various situations demand different ones of these techniques; we will provide automated methods for selecting repair methods in a situationally appropriate manner.

Many domains dictate that planning systems should be tools for aiding humans to generate plans that meet their needs rather than stand-alone software packages. For this reason, the CPEF will include a significant user guidance module that will enable users to influence and direct the planning and plan revision processes as they see fit. This module will be grounded in the metaphor of advice-taking, thus providing a natural and intuitive mode of interaction for users.

Core Technologies

The project will leverage a number of AI technologies developed previously at SRI.

Demonstrations

Publications


Last modified: October, 1999