Action languages and domain modeling

Richard Glenn Watson, University of Texas at El Paso

Abstract

We live in a dynamic world, full of situations which can be manipulated by our actions. The ability to give formal models of these dynamic domains is useful, whether to provide tools for humans in complex situations or a means of reasoning for robots in such domains. In recent years, there has been a considerable amount of research into action languages and their use for domain modeling. Such languages allow the user to describe the effects of actions in concise, easy to read statements. For computation, these statements can be translated into a logic program and queried. In many cases, the programs produced by such translations yield sound and complete answers to such queries. ^ In this dissertation we present an action language, L0. The language includes many standard features of action languages: static and dynamic causal laws, executability conditions, initial situation axioms, and queries. In this language we also introduce a new type of statements called definition propositions. Definition propositions are a simple yet elegant shorthand for sets of static causal laws. They thereby allow the user to create shorter domain descriptions. ^ We also present a translation of domains written using L0 into extended logic programs. This translation takes advantage of the use of definition propositions, leading to more efficient query answering under Prolog interpreters. ^ We prove that, for large classes of domain descriptions, the answers computed by a Prolog interpreter will be correct with respect to the domain. We further show how, using these results, the system can be used to verify plans to achieve goals within such domains. ^ We use these ideas for the development of a decision support system for the Space Shuttle. In particular, we represent knowledge about the Reaction Control System as a domain description of L0. We use its logic programming translation to implement a simple yet efficient plan checking system. The same domain description is used, together with CCALC to find plans for control of the RCS in multiple failure situations. ^

Subject Area

Computer Science

Recommended Citation

Watson, Richard Glenn, "Action languages and domain modeling" (1999). ETD Collection for University of Texas, El Paso. AAI9947588.
http://digitalcommons.utep.edu/dissertations/AAI9947588

Share

COinS