Admissible Heuristics for Optimal Planning

tlSP ~tm[ HSPr iLrt~ two r(’¢’Pllt p].allllors th~tt st.,trch th," statt~-Sl);.’," 1tsiltg i).lt heuristic fl|n,’tiolt (’xtt’act,’,l froill Strips cnt’tMiatgs, liSP dot’s a f(,rward s(mrch frtmt the ialt.bd stlttt, rt:vc,tuputiltg tim Imuristic in vv(.ry static. whih: HSPr tit)its a l’t:grt.ssit,u s(.~Lr0’h fl’cJm th(. g0,:tl t’,,utpuling ~t .-uitalth. r(.prv.~vtttati,)u -f th(. h,.,trim.it .nly om.v. Bt)th plmmvrs have .sht,wn Ko, td pvrfi)rman,’o. .ftt.n lm-lu,’ing st dutic uts that a.rt. t’()lnpot it iv(" in tim(" mM numl)vr of at:ti, ms with the s.hlti(,us fi,m,d Graphld~at nd SAT planno.rs. HSP and HSPr. h,,wvvt.r. ar,~ uot optimal plaonrrs. This is Imcmts(. tht. |t(.uristit’ fu|,cti, m is n.t ~ulntissil,h: aml tit(. s,.arch Mg,,rithnts arv not ,,ptintal. In this palm/we addrvss this probh.m. ’~,Vc fi)rntulatv ,t now ,tthnissiblc Iwuristlc fin’ lflanuing. us(~ it m guido :ut IDA* s(.art:h, a|ttl ,.mpirically (.valuate tit," result.trig optimM plann,’r ov(.r ~, llllllll)(.r dt nue.hta. Tit(’ maiu c.ntrilmti,,n is tit(. id,’a un, h.riyivg tht" Imuristic th~Lt yivhIs n.t (tilt" 11111". i’t whuh~ fim,ily of Imlyutmlial att,I a, huissild," houristi,’s that Ira, b. avt:uracy fitr (’fiici(.ncy. Tim formulation is gt’not’~d and sheds s.me light .n the hvuristics usvd ill liSP and Gr~qfltplan. and their rohttitm. It cxphfits tim fact,,rvd [Strips) rc.lm’s(’ntat.i~m c~f phtmting pr.bh.nm, mapping sh,rtcst.-imt.h probh:ms in st,,tr-space into suit~tbly (l’r-fln,’d sl.,rt(~st-patl, pr.l,hmls in otom-.~pacv. Tim fi)rvmlati-n applio.s with litt.lc vari:di-n to seqm.vtlal attd Itaxallvl l)[anning, and prtddcms will, ditll.rt.nt m:tiou ,.,,sts.