Automated Conversion and Simplification of Plan Representations
Martin Allen
Shlomo Zilberstein
Abstract
As planning agents grow more sophisticated, issues of plan
representation arise alongside concerns with plan generation.
Planning methods work over increasingly large and difficult
problems and resulting plans are often complex or unwieldy.
Further, where planners must interact with human beings--either
for purposes of plan verification and analysis, or in
mixed-initiative plan-generation settings--plans must be
represented
so that the intended course of action is readily visible.
We propose automated techniques for the simplification
of plans, and for conversion between distinct plan representations;
our proposal is illustrated by examples from our
recent research, concerning conversion between large-scale
MDP solutions and graph-based contingency plans.
Download
[pdf]