SP-POMDP：堆叠物体抓取场景中的任务规划方法

打开文本图片集

关键词：机器人抓取；堆叠场景；POMDP；任务规划；状态空间修正中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2025）07-019-2064-08doi：10.19734/j. issn.1001-3695.2024.11.0495

Abstract：Inthe workingscenarioof robots grasping stackedobjects，due tosensorsampling afectedbyclutter andpartial observabilitycaused byobjectocclusion intheenvironment，robotsareunable toachieveaccurateand complete modeling， making itdificulttoeficientlycompletetasks.ThispaperdesignedastatepatchedbasedpartiallyobservableMarkovdecision proceses （SP-POMDP）model to addressthe above isues，and proposed arobot grasping task planning method based on this model.This methodabstractly extracted discrete states，actions，andobservation spaces basedonsamplingresults.Through a statespacecoectionmethod，stateinformationthatcouldn’tbesampledandrecognizedduetopartialobservabilityintheenvironment wasadded tothestate spacebasedonthe inherentcharacteristicsofthecurent stackedscene.Itconstructedabelief treetosolvethemodel.Theresultsofexperimentsshowthatinthegraspingtaskofstackedobjects，thismethodcansignificantly reduce computation time and improve work eficiency while ensuring success rate.

Key words：robot grasping；stacking scene；POMDP；task planning；state space patching

0 引言

在杂乱的物体堆叠场景中搜索并抓取特定物体是工业场景中的场景任务。（剩余20310字）