Point　cloud　registration　is　a　crucial　link　in　robotic　arm　visual　grasping,　and　ICP　point　cloud　registration　algorithm　is　usually　used　to　obtain　the　pose　information　of　grasped　objects.　However,　the　current　algorithm　has　the　problems　of　long　registration　time　and　low　accuracy.　In　this　paper,　a　CPD-ICP　algorithm　is　proposed　for　the　grasping　environment　of　the　robotic　arm.　The　CPD　algorithm　is　used　as　the　rough　registration　to　obtain　the　initial　pose　of　the　source　point　cloud,　and　then　the　ICP　algorithm　is　used　for　precise　registration.　The　closed-loop　grasping　efficiency　and　end-end　accuracy　of　the　robotic　arm　are　verified　by　testing　experiments　on　public　datasets　such　as　Stanford　Bunny　and　the　TOOL-21　dataset　created　in　this　paper.　A　large　number　of　experiments　show　that　this　method　improves　the　grasping　efficiency　of　the　robotic　arm　and　the　grasping　point　at　the　end　of　the　robotic　arm　is　more　accurate.　©　2022　IEEE.