@TECHREPORT{R-95-12,
PSURL = {/publications/cgi-bin/tr-fetch.pl?r-95-12+ps},
NUMBER = {R-95-12},
INSTITUTION = ida,
ADDRESS = idaaddr,
YEAR = {1995},
AUTHOR = {B{\"a}ckstr{\"o}m, Christer and Jonsson, Peter},
TITLE = {PLANNING WITH ABSTRACTION HIERARCHIES CAN BE EXPONENTIALLY LESS EFFICIENT},
ABSTRACTURL = {/publications/cgi-bin/tr-fetch.pl?r-95-12+abstr},
ABSTRACT = {It is well-known that state abstraction can speed up planning exponentially, under ideal conditions. We add to the knowledge{\ae}showing that state abstraction may likewise slow down planning exponentially, and even result in generating an exponentially longer solution than necessary. This phenomenon can occur for abstraction hierarchies which are generated automatically by the ALPINE AND HIGHPOINT algorithms. We further show that there is little hope of any drastic improvement upon these algorithms{\ae}it is computationally difficult to generate abstraction hierarchies which allow finding good approximations of optimal plans.}