Date of Original Version



Technical Report

Rights Management

All Rights Reserved

Abstract or Description

Abstract: "Two important factors hinder our ability to address large planning problems. On one hand, our understanding of planning is not independent from specific planning frameworks. On the other hand, current planning frameworks lack modularity, a key factor for 'divide and conquer' approaches to large problems. This paper addresses the formal definition of planning, points out some limitations of the current planning frameworks, and describes a new planning framework that overcomes these limitations. Our formal definition relies on the hypothesis that the problem solver's model world is a dynamical system.On this basis, we can clearly separate the knowledge about how the world works from the heuristic knowledge needed to solve problems quickly. Our definition is also independent from any particular planning representation framework. Our analysis of modularity indicates which features can support it in a planning framework. We describe how these features are implemented in the HSTS planning framework, a general purpose facility integrated in the HSTS scheduling architecture. Its effectiveness to address complex 'real world' domains has been successfully demonstrated on the problem of building executable observation schedules for the Hubble Space Telescope."

Included in

Robotics Commons