The Breakdown of Operators When Interacting With the External World

By looking at the simple task of tossing a bean bag from hand to hand, we show how the macro operator method breaks down when formulating agent models that interact with an uncertain external world. A macro operator encapsulates a plan to reach an objective. Occasionally the objective will be found...

Full description

Saved in:
Bibliographic Details
Main Authors: Pelton, Garrett A, Lehman, Jill F
Format: Report
Language:English
Subjects:
ARCHITECTURE
ARTIFICIAL INTELLIGENCE
Cybernetics
ERRORS
EXTERNAL
FORMULATIONS
INTERACTIONS
LEARNING
PSCM(PROBLEM SPACE COMPUTATIONAL MODEL)
RECOVERY
REJECTION
SEQUENCES
TRANSFORMATIONS
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:By looking at the simple task of tossing a bean bag from hand to hand, we show how the macro operator method breaks down when formulating agent models that interact with an uncertain external world. A macro operator encapsulates a plan to reach an objective. Occasionally the objective will be found to be unachievable, requiring the macro operator and its plan to be rejected. Letting the macro operator interact with the external world does not, by itself, change this situation. but the fact that the results of the interaction are uncertain, and the agent's knowledge incomplete. does. The key idea is that the agent can't positively determine if progress towards the objective is being made in the external world, and thus errors will be made in rejecting a macro operator that would succeed. We show that there are a number of methods by which the agent can recover from such an operator rejection and continue toward the operator's objective. If we make operator rejection and recovery into a common mechanism, then the operators and the plans they represent will be split by the interaction into a sequence of smaller operators each doing a portion of the work toward the objective of the larger operator. The models are described in terms of Soar and we assume the reader's familiarity with both the architecture and the Problem Space Computational Model in our discussions. Artificial Intelligence, Learning, Plan formulation, Plan execution, Program transformation, Soar.