Accession Number:

ADA288873

Title:

Multi-Method Planning.

Descriptive Note:

Research rept.,

Corporate Author:

UNIVERSITY OF SOUTHERN CALIFORNIA MARINA DEL REY INFORMATION SCIENCES INST

Personal Author(s):

• Lee, Soowon

Report Date:

1994-04-01

Pagination or Media Count:

156.0

Abstract:

The ability to find a low execution-cost plan efficiently over a wide domain of applicability is the core of domain-independent planning systems. The approach investigated here to building such a planning system begins with two hypotheses 1 no single method will satisfy both sufficiency and efficiency for all situations and 2 multi-method planning can out-perform single-method planning in terms of sufficiency and efficiency. To evaluate these hypotheses, a set of single-method planners for the domains investigated show that these planners have trouble performing efficiently over a wide range of problems. As an alternative to single-method planning, multi-method planning is investigated in this thesis. A multi-method planner consists of a coordinated set of methods which have different performance and scope. Given a set of created methods, the key issue in multi-method planning is how to coordinate individual methods in an efficient manner so that the multi-method plainer can have high performance. The multi-method framework presented here provides one way to do this basset on the notion of bias-relaxation. In a bias-relaxation multi-method planner, planning starts by trying highly restricted and efficient methods, and then successively relaxes restrictions until a sufficient method is found.

Descriptors:

• *LEARNING MACHINES
• *ARTIFICIAL INTELLIGENCE
• EFFICIENCY
• THESES
• LIMITATIONS
• PLANNING
• HYPOTHESES
• RANGE(EXTREMES)

Subject Categories:

• Psychology
• Cybernetics

Distribution Statement:

APPROVED FOR PUBLIC RELEASE