基于案例推理的路径规划分析-path planning analysis based on case-based reasoning.docxVIP

摘要摘要路径搜索是计算机游戏中最为常见的任务之一，搜索算法的质量很大程度上影响着游戏的趣味性与可玩性。A*算法是最典型的启发式搜索算法，在路径确实存在的情况下，它能够确保得到一条最优路径。然而，该算法有着较高的时间和空间复杂度，因此，它不适用于多任务快速路径规划的问题。为了使A*算法满足多任务快速路径规划的需求，本文提出了基于案例推理的路径规划算法。采用离线和在线过程相结合，提出了案例（即已知路径的起始终止点对）的kd 树存储和查询机制，并给出了案例的重用方法。首先，在离线状态下，随机产生和保存一些路径作为已知案例，并根据这些案例构建kd 树。其次，在线状态下，当新任务出现的时候，我们不再简单的利用A*算法从头进行搜索，而是在kd树中快速查询最相似的案例。然后我们给出了案例的重用机制：判断所查询到的这个案例是否满足预先设定的阈值。如果满足阈值条件，那么我们就利用这条所选出来的预存相似案例，然后对该路径进行适当的调整来得到新任务的解决方案。否则，我们就要利用A*对这个新任务进行路径规划。最终，该算法通过牺牲部分内存空间来存储案例，缩短了路径规划的时间。试验结果表明，在特定的存储条件下，随着存储案例的增多，新任务能够取得更高效的路径规划。关键词K维空间树最近邻算法基于案例的推理曼哈顿距离ⅠAbstractAbstractPathfindingisoneofthemostcommontasksincomputergames,anditsqualityhasagreatinfluenceupontheinterestofGame.A*isofthemosttypicalheuristicsearchalgorithm, whichcanfindanoptimalpathifsuchapathreallyexists.However,A*suffersfromhightimeandspacecomplexity,therefore,itisnotsuitableforthemulti-taskfastpathfinding.In ordertomakeA*tomeettheneedsofmulti-taskfastpathfinding,weintroducedcase-basedreasoningmethodintheprocessofA*algorithminthispaper.Combiningtheprocessof onlineandoffline,weputforwardakindofkdtree’storingandqueryingmechanismofthecase(s)andprovidethecase(s)reusingmethod.Firstly,wegeneratesomepathsandstorethemascasesandbuildakd-treeaccordingtothecasesinthecasebaseoffline.Secondly,whenanewtaskiscomingonline,itnolongerusesA*tofindapathfromscratch,butfirstlyquerythemostsimilarcaseinthekd-tree.Then,weprovidethereusingmechanismofthecase(s):judgetheselectedcasewhethermeetthethresholdconditionwesetinadvanceornot.Iftheselectedcasemeets the thresholdconditionwesetbefore,thenwewillreusetheselectedpathwhichispre-storedinthecasebase.Asolutiontothenewtaskwillbeobtainedafteradaptingthefoundsimilarcase(s).Otherwise,thepathfindingwillbedonebyA*independently.Finally,thealgorithmcanreducethesearchtimeatthecostofusingmorememorytostorefoundpaths.Theexperimentalresultsshowthat,undercertainstorageconditions,withtheincreasingnumberofthestoredpaths,thenewcomingtaskwillhaveanefficientimplementation.KeywordsKd-treeNearest-neighborssearcha