Motivation

Many domains in which AI planning techniques can be profitably employed are dynamic in nature. For example, military operations planning and controlling a mobile robot both exhibit this characteristic: during either plan generation or plan execution, the state of the world can change dramatically as troops are dispatched to an area or a robot navigates through a hallway. For such domains, it is necessary that plan generation systems be sensitive to run-time concerns and that plan execution systems be capable of invoking the plan generator to address unexpected events at run-time.

Traditionally, generative planning and plan execution have been considered distinct activities. This dichotomy has lead to the development of generative planning systems that ignore execution issues and reactive plan execution systems that cannot synthesize new plans at run-time. Not surprisingly, different representations have developed for defining operators for synthesizing and executing plans, making it difficult to support tightly integrated plan generation and plan execution systems.

Overview

SRI International has developed a framework called the Act formalism for expressing the kind of procedural information required by both plan-generation and plan-execution systems. At the heart of the formalism is the Act structure, which corresponds to both an operator or plan in the terminology of generative planners, and a plan fragment or operating procedure in the terminology of plan execution systems. Act structures (also referred to as Acts) provide a common representation language that bridges the gap between these two types of plan-based systems. Development of the Act formalism was motivated by experience in combining a generative planning system, SIPE-2, and a reactive plan-execution system PRS-CL, to form the basis of the integrated planning environment, Cypress. While the details of Acts are tailored to these two specific planning systems, the underlying philosophy of Acts applies more generally to many types of planning formalisms.

The online postcript document The Act Formalism provides a brief overview of the Act formalism, including a BNF grammar specification of Act. Additional details on the semantics and backgound on Act can be found in the references provided below.

The Act Editor

The Act-Editor is a graphical browsing and editing system for procedural knowledge expressed as Acts. Through it, Acts can be created, viewed, and edited through direct picotiral manipulation.

References

K. L. Myers and D. E. Wilkins,
"The Act Formalism", Version 2.2,
SRI International Artificial Intelligence Center, Menlo Park, CA, September 1997.
Abstract: Provides a brief overview of the Act formalism, including a BNF grammar specification of Act.

D. E. Wilkins and K. L. Myers,
"A common knowledge representation for plan generation and reactive execution,"
Journal of Logic and Computation, vol. 5, number 6, pp. 731--761, December 1995.

Abstract: The ability to integrate sophisticated planning techniques with reactive execution systems is critical for nontrivial applications. Merging these two technologies is difficult because the forms of knowledge and reasoning that they employ differ substantially. The ACT formalism is a language for representing the knowledge required to support both the generation of complex plans and reactive execution of those plans in dynamic environments. A design goal of ACT was its adequacy for practical applications. ACT has been used as the interlingua in an implemented system that links a previously implemented planner with a previously implemented executor. This system has been used in several applications, including robot control and military operations, thus attesting to its expressive and computational adequacy.


Back to Karen L. Myers, <myers@ai.sri.com>
Back to David E. Wilkins, <wilkins@ai.sri.com>
Back to Artificial Intelligence Center
Back to SRI International

Act-Editor, Cypress, Grasper-CL, PRS-CL, and SIPE-2 are trademarks of SRI International.
Copyright © 1995-2003, SRI International, 333 Ravenswood Ave., Menlo Park, CA 94025 USA. All rights reserved.