With　the　aging　of　the　population　and　the　increase　in　the　employment　cost　of　the　elderly　service　industry,　the　elderly　service　robot　has　become　an　important　development　field.　However,　the　working　environment　of　the　service　robot　is　complex,　and　the　robot　arm　may　collide　with　the　environmental　obstacles,　which　will　cause　damage　not　only　to　the　obstacles　but　also　to　the　robot　arm　and　even　the　served　object.　In　this　paper,　an　improved　rapidly　exploring　random　tree　algorithm　is　proposed　for　path　planning　of　humanoid　robotic　arm.　First,　the　threshold　of　random　probability　sampling　is　set　as　the　selection　criterion　of　the　extension　mode.　Then　the　random　sampling　value　and　the　probability　sampling　threshold　are　compared,　and　the　target　point　is　determined　as　a　random　point　if　the　random　sampling　value　is　within　the　set　threshold　value.　Otherwise,　the　new　node　will　be　expanded　under　the　joint　action　of　the　target　gravity　and　the　random　point　gravity.　This　is　the　extension　of　classification　based　on　random　probability.　Finally,　the　breadth-first　search　is　used　to　smooth　the　initial　exploration　path.　The　simulation　results　show　that　the　proposed　method　can　effectively　reduce　the　path　planning　time　and　path　length.　©　2020　IEEE.