id sid tid token lemma pos fcis-28493 1 1 frontiers frontier NOUN fcis-28493 1 2 in in ADP fcis-28493 1 3 computing computing NOUN fcis-28493 1 4 and and CCONJ fcis-28493 1 5 intelligent intelligent ADJ fcis-28493 1 6 systems system NOUN fcis-28493 1 7 issn issn VERB fcis-28493 1 8 : : PUNCT fcis-28493 1 9 2832 2832 NUM fcis-28493 1 10 - - SYM fcis-28493 1 11 6024 6024 NUM fcis-28493 1 12 | | NOUN fcis-28493 1 13 vol vol NOUN fcis-28493 1 14 . . PROPN fcis-28493 2 1 10 10 NUM fcis-28493 2 2 , , PUNCT fcis-28493 2 3 no no INTJ fcis-28493 2 4 . . NOUN fcis-28493 2 5 3 3 NUM fcis-28493 2 6 , , PUNCT fcis-28493 2 7 2024 2024 NUM fcis-28493 2 8 65 65 NUM fcis-28493 2 9 research research NOUN fcis-28493 2 10 on on ADP fcis-28493 2 11 robotic robotic ADJ fcis-28493 2 12 arm arm NOUN fcis-28493 2 13 grasping grasp VERB fcis-28493 2 14 algorithm algorithm NOUN fcis-28493 2 15 based base VERB fcis-28493 2 16 on on ADP fcis-28493 2 17 an an DET fcis-28493 2 18 enhanced enhanced ADJ fcis-28493 2 19 edge edge NOUN fcis-28493 2 20 network network PROPN fcis-28493 2 21 hongyi hongyi PROPN fcis-28493 2 22 li li PROPN fcis-28493 2 23 , , PUNCT fcis-28493 2 24 jianjun jianjun PROPN fcis-28493 2 25 zhu zhu PROPN fcis-28493 2 26 school school NOUN fcis-28493 2 27 of of ADP fcis-28493 2 28 information information NOUN fcis-28493 2 29 and and CCONJ fcis-28493 2 30 control control PROPN fcis-28493 2 31 engineering engineering PROPN fcis-28493 2 32 , , PUNCT fcis-28493 2 33 jilin jilin PROPN fcis-28493 2 34 institute institute PROPN fcis-28493 2 35 of of ADP fcis-28493 2 36 chemical chemical PROPN fcis-28493 2 37 technology technology PROPN fcis-28493 2 38 , , PUNCT fcis-28493 2 39 jilin jilin PROPN fcis-28493 2 40 , , PUNCT fcis-28493 2 41 jilin jilin PROPN fcis-28493 2 42 132022 132022 PROPN fcis-28493 2 43 , , PUNCT fcis-28493 2 44 china china PROPN fcis-28493 2 45 abstract abstract NOUN fcis-28493 2 46 : : PUNCT fcis-28493 2 47 to to PART fcis-28493 2 48 address address VERB fcis-28493 2 49 the the DET fcis-28493 2 50 issue issue NOUN fcis-28493 2 51 of of ADP fcis-28493 2 52 low low ADJ fcis-28493 2 53 grasp grasp NOUN fcis-28493 2 54 detection detection NOUN fcis-28493 2 55 accuracy accuracy NOUN fcis-28493 2 56 in in ADP fcis-28493 2 57 robotic robotic ADJ fcis-28493 2 58 arms arm NOUN fcis-28493 2 59 , , PUNCT fcis-28493 2 60 an an DET fcis-28493 2 61 improved improved ADJ fcis-28493 2 62 point point NOUN fcis-28493 2 63 cloud cloud NOUN fcis-28493 2 64 - - PUNCT fcis-28493 2 65 based base VERB fcis-28493 2 66 grasp grasp NOUN fcis-28493 2 67 detection detection NOUN fcis-28493 2 68 model model NOUN fcis-28493 2 69 , , PUNCT fcis-28493 2 70 tes tes NOUN fcis-28493 2 71 - - NOUN fcis-28493 2 72 net net ADJ fcis-28493 2 73 ( ( PUNCT fcis-28493 2 74 transformedgesage transformedgesage NOUN fcis-28493 2 75 network network NOUN fcis-28493 2 76 ) ) PUNCT fcis-28493 2 77 , , PUNCT fcis-28493 2 78 is be AUX fcis-28493 2 79 proposed propose VERB fcis-28493 2 80 , , PUNCT fcis-28493 2 81 which which PRON fcis-28493 2 82 builds build VERB fcis-28493 2 83 upon upon SCONJ fcis-28493 2 84 the the DET fcis-28493 2 85 edge edge NOUN fcis-28493 2 86 grasp grasp NOUN fcis-28493 2 87 network network NOUN fcis-28493 2 88 architecture architecture NOUN fcis-28493 2 89 . . PUNCT fcis-28493 3 1 in in ADP fcis-28493 3 2 this this DET fcis-28493 3 3 model model NOUN fcis-28493 3 4 , , PUNCT fcis-28493 3 5 pointtransformerconv pointtransformerconv ADJ fcis-28493 3 6 layers layer NOUN fcis-28493 3 7 are be AUX fcis-28493 3 8 first first ADV fcis-28493 3 9 employed employ VERB fcis-28493 3 10 to to PART fcis-28493 3 11 extract extract VERB fcis-28493 3 12 local local ADJ fcis-28493 3 13 features feature NOUN fcis-28493 3 14 from from ADP fcis-28493 3 15 the the DET fcis-28493 3 16 point point NOUN fcis-28493 3 17 cloud cloud NOUN fcis-28493 3 18 , , PUNCT fcis-28493 3 19 integrating integrate VERB fcis-28493 3 20 a a DET fcis-28493 3 21 self self NOUN fcis-28493 3 22 - - PUNCT fcis-28493 3 23 attention attention NOUN fcis-28493 3 24 mechanism mechanism NOUN fcis-28493 3 25 to to PART fcis-28493 3 26 capture capture VERB fcis-28493 3 27 the the DET fcis-28493 3 28 complex complex ADJ fcis-28493 3 29 relationships relationship NOUN fcis-28493 3 30 inherent inherent ADJ fcis-28493 3 31 within within ADP fcis-28493 3 32 the the DET fcis-28493 3 33 point point NOUN fcis-28493 3 34 cloud cloud NOUN fcis-28493 3 35 data datum NOUN fcis-28493 3 36 . . PUNCT fcis-28493 4 1 subsequently subsequently ADV fcis-28493 4 2 , , PUNCT fcis-28493 4 3 the the DET fcis-28493 4 4 sageconv sageconv NOUN fcis-28493 4 5 graph graph NOUN fcis-28493 4 6 neural neural ADJ fcis-28493 4 7 network network NOUN fcis-28493 4 8 module module NOUN fcis-28493 4 9 is be AUX fcis-28493 4 10 utilized utilize VERB fcis-28493 4 11 to to PART fcis-28493 4 12 perform perform VERB fcis-28493 4 13 feature feature NOUN fcis-28493 4 14 aggregation aggregation NOUN fcis-28493 4 15 on on ADP fcis-28493 4 16 the the DET fcis-28493 4 17 adjacency adjacency NOUN fcis-28493 4 18 graph graph NOUN fcis-28493 4 19 , , PUNCT fcis-28493 4 20 thereby thereby ADV fcis-28493 4 21 mitigating mitigate VERB fcis-28493 4 22 the the DET fcis-28493 4 23 issue issue NOUN fcis-28493 4 24 of of ADP fcis-28493 4 25 neglecting neglect VERB fcis-28493 4 26 the the DET fcis-28493 4 27 inter inter ADJ fcis-28493 4 28 - - ADJ fcis-28493 4 29 point point ADJ fcis-28493 4 30 relationships relationship NOUN fcis-28493 4 31 during during ADP fcis-28493 4 32 the the DET fcis-28493 4 33 feature feature NOUN fcis-28493 4 34 extraction extraction NOUN fcis-28493 4 35 phase phase NOUN fcis-28493 4 36 and and CCONJ fcis-28493 4 37 enhancing enhance VERB fcis-28493 4 38 the the DET fcis-28493 4 39 overall overall ADJ fcis-28493 4 40 network network NOUN fcis-28493 4 41 performance performance NOUN fcis-28493 4 42 . . PUNCT fcis-28493 5 1 experimental experimental ADJ fcis-28493 5 2 results result NOUN fcis-28493 5 3 demonstrate demonstrate VERB fcis-28493 5 4 that that SCONJ fcis-28493 5 5 , , PUNCT fcis-28493 5 6 in in ADP fcis-28493 5 7 comparison comparison NOUN fcis-28493 5 8 to to ADP fcis-28493 5 9 existing exist VERB fcis-28493 5 10 state state NOUN fcis-28493 5 11 - - PUNCT fcis-28493 5 12 of of ADP fcis-28493 5 13 - - PUNCT fcis-28493 5 14 the the DET fcis-28493 5 15 - - PUNCT fcis-28493 5 16 art art NOUN fcis-28493 5 17 point point NOUN fcis-28493 5 18 cloud cloud NOUN fcis-28493 5 19 grasp grasp NOUN fcis-28493 5 20 detection detection NOUN fcis-28493 5 21 methods method NOUN fcis-28493 5 22 , , PUNCT fcis-28493 5 23 the the DET fcis-28493 5 24 proposed propose VERB fcis-28493 5 25 model model NOUN fcis-28493 5 26 exhibits exhibit VERB fcis-28493 5 27 superior superior ADJ fcis-28493 5 28 generalization generalization NOUN fcis-28493 5 29 capability capability NOUN fcis-28493 5 30 , , PUNCT fcis-28493 5 31 improved improve VERB fcis-28493 5 32 robustness robustness NOUN fcis-28493 5 33 and and CCONJ fcis-28493 5 34 stability stability NOUN fcis-28493 5 35 , , PUNCT fcis-28493 5 36 as as ADV fcis-28493 5 37 well well ADV fcis-28493 5 38 as as ADP fcis-28493 5 39 higher high ADJ fcis-28493 5 40 accuracy accuracy NOUN fcis-28493 5 41 . . PUNCT fcis-28493 6 1 keywords keyword NOUN fcis-28493 6 2 : : PUNCT fcis-28493 6 3 robotic robotic ADJ fcis-28493 6 4 arm arm NOUN fcis-28493 6 5 grasp grasp NOUN fcis-28493 6 6 detection detection NOUN fcis-28493 6 7 ; ; PUNCT fcis-28493 6 8 edge edge NOUN fcis-28493 6 9 grasp grasp NOUN fcis-28493 6 10 detection detection NOUN fcis-28493 6 11 network network NOUN fcis-28493 6 12 ; ; PUNCT fcis-28493 6 13 self self NOUN fcis-28493 6 14 - - PUNCT fcis-28493 6 15 attention attention NOUN fcis-28493 6 16 mechanism mechanism NOUN fcis-28493 6 17 ; ; PUNCT fcis-28493 6 18 graph graph NOUN fcis-28493 6 19 neural neural ADJ fcis-28493 6 20 network network NOUN fcis-28493 6 21 module module NOUN fcis-28493 6 22 . . PUNCT fcis-28493 7 1 1 1 X fcis-28493 7 2 . . X fcis-28493 7 3 introduction introduction NOUN fcis-28493 7 4 in in ADP fcis-28493 7 5 recent recent ADJ fcis-28493 7 6 years year NOUN fcis-28493 7 7 , , PUNCT fcis-28493 7 8 robotic robotic ADJ fcis-28493 7 9 arm arm NOUN fcis-28493 7 10 grasping grasping NOUN fcis-28493 7 11 has have AUX fcis-28493 7 12 gradually gradually ADV fcis-28493 7 13 transitioned transition VERB fcis-28493 7 14 from from ADP fcis-28493 7 15 repetitive repetitive ADJ fcis-28493 7 16 tasks task NOUN fcis-28493 7 17 to to ADP fcis-28493 7 18 more more ADV fcis-28493 7 19 intelligent intelligent ADJ fcis-28493 7 20 operations operation NOUN fcis-28493 7 21 , , PUNCT fcis-28493 7 22 necessitating necessitate VERB fcis-28493 7 23 higher high ADJ fcis-28493 7 24 demands demand NOUN fcis-28493 7 25 on on ADP fcis-28493 7 26 the the DET fcis-28493 7 27 system system NOUN fcis-28493 7 28 , , PUNCT fcis-28493 7 29 which which PRON fcis-28493 7 30 must must AUX fcis-28493 7 31 be be AUX fcis-28493 7 32 capable capable ADJ fcis-28493 7 33 of of ADP fcis-28493 7 34 computing compute VERB fcis-28493 7 35 multiple multiple ADJ fcis-28493 7 36 potential potential ADJ fcis-28493 7 37 grasp grasp NOUN fcis-28493 7 38 poses pose NOUN fcis-28493 7 39 . . PUNCT fcis-28493 8 1 at at ADP fcis-28493 8 2 present present ADJ fcis-28493 8 3 , , PUNCT fcis-28493 8 4 mainstream mainstream ADJ fcis-28493 8 5 methods method NOUN fcis-28493 8 6 for for ADP fcis-28493 8 7 robotic robotic ADJ fcis-28493 8 8 arm arm NOUN fcis-28493 8 9 grasp grasp NOUN fcis-28493 8 10 pose pose NOUN fcis-28493 8 11 detection detection NOUN fcis-28493 8 12 predominantly predominantly ADV fcis-28493 8 13 rely rely VERB fcis-28493 8 14 on on ADP fcis-28493 8 15 2d 2d NUM fcis-28493 8 16 images image NOUN fcis-28493 8 17 . . PUNCT fcis-28493 9 1 however however ADV fcis-28493 9 2 , , PUNCT fcis-28493 9 3 these these DET fcis-28493 9 4 methods method NOUN fcis-28493 9 5 are be AUX fcis-28493 9 6 limited limit VERB fcis-28493 9 7 by by ADP fcis-28493 9 8 their their PRON fcis-28493 9 9 ability ability NOUN fcis-28493 9 10 to to PART fcis-28493 9 11 only only ADV fcis-28493 9 12 provide provide VERB fcis-28493 9 13 projection projection NOUN fcis-28493 9 14 information information NOUN fcis-28493 9 15 of of ADP fcis-28493 9 16 objects object NOUN fcis-28493 9 17 onto onto ADP fcis-28493 9 18 a a DET fcis-28493 9 19 2d 2d NUM fcis-28493 9 20 plane plane NOUN fcis-28493 9 21 , , PUNCT fcis-28493 9 22 lacking lack VERB fcis-28493 9 23 depth depth NOUN fcis-28493 9 24 and and CCONJ fcis-28493 9 25 stereoscopic stereoscopic ADJ fcis-28493 9 26 perception perception NOUN fcis-28493 9 27 . . PUNCT fcis-28493 10 1 consequently consequently ADV fcis-28493 10 2 , , PUNCT fcis-28493 10 3 they they PRON fcis-28493 10 4 are be AUX fcis-28493 10 5 less less ADV fcis-28493 10 6 effective effective ADJ fcis-28493 10 7 in in ADP fcis-28493 10 8 handling handle VERB fcis-28493 10 9 complex complex ADJ fcis-28493 10 10 scenes scene NOUN fcis-28493 10 11 and and CCONJ fcis-28493 10 12 object object VERB fcis-28493 10 13 poses pose NOUN fcis-28493 10 14 . . PUNCT fcis-28493 11 1 in in ADP fcis-28493 11 2 contrast contrast NOUN fcis-28493 11 3 , , PUNCT fcis-28493 11 4 point point VERB fcis-28493 11 5 cloud cloud NOUN fcis-28493 11 6 data datum NOUN fcis-28493 11 7 directly directly ADV fcis-28493 11 8 captures capture VERB fcis-28493 11 9 the the DET fcis-28493 11 10 position position NOUN fcis-28493 11 11 and and CCONJ fcis-28493 11 12 shape shape NOUN fcis-28493 11 13 of of ADP fcis-28493 11 14 objects object NOUN fcis-28493 11 15 in in ADP fcis-28493 11 16 three three NUM fcis-28493 11 17 - - PUNCT fcis-28493 11 18 dimensional dimensional ADJ fcis-28493 11 19 space space NOUN fcis-28493 11 20 , , PUNCT fcis-28493 11 21 offering offer VERB fcis-28493 11 22 rich rich ADJ fcis-28493 11 23 depth depth NOUN fcis-28493 11 24 information information NOUN fcis-28493 11 25 . . PUNCT fcis-28493 12 1 this this PRON fcis-28493 12 2 enables enable VERB fcis-28493 12 3 more more ADV fcis-28493 12 4 accurate accurate ADJ fcis-28493 12 5 determination determination NOUN fcis-28493 12 6 of of ADP fcis-28493 12 7 an an DET fcis-28493 12 8 object object NOUN fcis-28493 12 9 ’s ’s PART fcis-28493 12 10 location location NOUN fcis-28493 12 11 , , PUNCT fcis-28493 12 12 orientation orientation NOUN fcis-28493 12 13 , , PUNCT fcis-28493 12 14 and and CCONJ fcis-28493 12 15 pose pose VERB fcis-28493 12 16 . . PUNCT fcis-28493 13 1 as as ADP fcis-28493 13 2 a a DET fcis-28493 13 3 result result NOUN fcis-28493 13 4 , , PUNCT fcis-28493 13 5 point point NOUN fcis-28493 13 6 cloud cloud NOUN fcis-28493 13 7 data datum NOUN fcis-28493 13 8 has have AUX fcis-28493 13 9 garnered garner VERB fcis-28493 13 10 increasing increase VERB fcis-28493 13 11 attention attention NOUN fcis-28493 13 12 and and CCONJ fcis-28493 13 13 application application NOUN fcis-28493 13 14 in in ADP fcis-28493 13 15 research research NOUN fcis-28493 13 16 that that PRON fcis-28493 13 17 requires require VERB fcis-28493 13 18 precise precise ADJ fcis-28493 13 19 grasp grasp NOUN fcis-28493 13 20 localization localization NOUN fcis-28493 13 21 in in ADP fcis-28493 13 22 recent recent ADJ fcis-28493 13 23 years year NOUN fcis-28493 13 24 . . PUNCT fcis-28493 14 1 in in ADP fcis-28493 14 2 the the DET fcis-28493 14 3 field field NOUN fcis-28493 14 4 of of ADP fcis-28493 14 5 robotics robotic NOUN fcis-28493 14 6 , , PUNCT fcis-28493 14 7 grasp grasp NOUN fcis-28493 14 8 detection detection NOUN fcis-28493 14 9 is be AUX fcis-28493 14 10 crucial crucial ADJ fcis-28493 14 11 . . PUNCT fcis-28493 15 1 robots robot NOUN fcis-28493 15 2 need need VERB fcis-28493 15 3 to to PART fcis-28493 15 4 identify identify VERB fcis-28493 15 5 a a DET fcis-28493 15 6 set set NOUN fcis-28493 15 7 of of ADP fcis-28493 15 8 feasible feasible ADJ fcis-28493 15 9 grasping grasping NOUN fcis-28493 15 10 poses pose NOUN fcis-28493 15 11 by by ADP fcis-28493 15 12 analyzing analyze VERB fcis-28493 15 13 images image NOUN fcis-28493 15 14 , , PUNCT fcis-28493 15 15 voxels voxel NOUN fcis-28493 15 16 , , PUNCT fcis-28493 15 17 or or CCONJ fcis-28493 15 18 point point VERB fcis-28493 15 19 clouds cloud NOUN fcis-28493 15 20 to to PART fcis-28493 15 21 achieve achieve VERB fcis-28493 15 22 stable stable ADJ fcis-28493 15 23 object object NOUN fcis-28493 15 24 manipulation manipulation NOUN fcis-28493 15 25 . . PUNCT fcis-28493 16 1 typically typically ADV fcis-28493 16 2 , , PUNCT fcis-28493 16 3 these these DET fcis-28493 16 4 methods method NOUN fcis-28493 16 5 are be AUX fcis-28493 16 6 categorized categorize VERB fcis-28493 16 7 into into ADP fcis-28493 16 8 two two NUM fcis-28493 16 9 groups group NOUN fcis-28493 16 10 : : PUNCT fcis-28493 16 11 se(2 se(2 NOUN fcis-28493 16 12 ) ) PUNCT fcis-28493 16 13 ( ( PUNCT fcis-28493 16 14 the the DET fcis-28493 16 15 special special ADJ fcis-28493 16 16 euclidean euclidean ADJ fcis-28493 16 17 group group NOUN fcis-28493 16 18 in in ADP fcis-28493 16 19 two two NUM fcis-28493 16 20 dimensions dimension NOUN fcis-28493 16 21 ) ) PUNCT fcis-28493 16 22 and and CCONJ fcis-28493 16 23 se(3 se(3 NOUN fcis-28493 16 24 ) ) PUNCT fcis-28493 16 25 ( ( PUNCT fcis-28493 16 26 the the DET fcis-28493 16 27 special special ADJ fcis-28493 16 28 euclidean euclidean ADJ fcis-28493 16 29 group group NOUN fcis-28493 16 30 in in ADP fcis-28493 16 31 three three NUM fcis-28493 16 32 dimensions)[1 dimensions)[1 NOUN fcis-28493 16 33 ] ] PUNCT fcis-28493 16 34 . . PUNCT fcis-28493 17 1 the the DET fcis-28493 17 2 se(2 se(2 NOUN fcis-28493 17 3 ) ) PUNCT fcis-28493 17 4 methods method NOUN fcis-28493 17 5 rely rely VERB fcis-28493 17 6 on on ADP fcis-28493 17 7 top top ADJ fcis-28493 17 8 - - PUNCT fcis-28493 17 9 view view NOUN fcis-28493 17 10 images image NOUN fcis-28493 17 11 for for ADP fcis-28493 17 12 inference inference NOUN fcis-28493 17 13 , , PUNCT fcis-28493 17 14 while while SCONJ fcis-28493 17 15 the the DET fcis-28493 17 16 se(3 se(3 NOUN fcis-28493 17 17 ) ) PUNCT fcis-28493 17 18 approaches approach NOUN fcis-28493 17 19 utilize utilize VERB fcis-28493 17 20 point point NOUN fcis-28493 17 21 clouds cloud NOUN fcis-28493 17 22 or or CCONJ fcis-28493 17 23 voxel voxel NOUN fcis-28493 17 24 grids grid NOUN fcis-28493 17 25 for for ADP fcis-28493 17 26 their their PRON fcis-28493 17 27 analysis analysis NOUN fcis-28493 17 28 . . PUNCT fcis-28493 18 1 compared compare VERB fcis-28493 18 2 to to ADP fcis-28493 18 3 se(2 se(2 NOUN fcis-28493 18 4 ) ) PUNCT fcis-28493 18 5 methods method NOUN fcis-28493 18 6 , , PUNCT fcis-28493 18 7 se(3 se(3 NOUN fcis-28493 18 8 ) ) PUNCT fcis-28493 18 9 techniques technique NOUN fcis-28493 18 10 offer offer VERB fcis-28493 18 11 significant significant ADJ fcis-28493 18 12 advantages advantage NOUN fcis-28493 18 13 as as SCONJ fcis-28493 18 14 they they PRON fcis-28493 18 15 can can AUX fcis-28493 18 16 more more ADV fcis-28493 18 17 readily readily ADV fcis-28493 18 18 identify identify VERB fcis-28493 18 19 side side NOUN fcis-28493 18 20 grasping grasp VERB fcis-28493 18 21 poses pose NOUN fcis-28493 18 22 of of ADP fcis-28493 18 23 objects object NOUN fcis-28493 18 24 . . PUNCT fcis-28493 19 1 by by ADP fcis-28493 19 2 processing processing NOUN fcis-28493 19 3 point point NOUN fcis-28493 19 4 cloud cloud NOUN fcis-28493 19 5 data datum NOUN fcis-28493 19 6 , , PUNCT fcis-28493 19 7 se(3 se(3 NOUN fcis-28493 19 8 ) ) PUNCT fcis-28493 19 9 methods method NOUN fcis-28493 19 10 comprehensively comprehensively ADV fcis-28493 19 11 account account VERB fcis-28493 19 12 for for ADP fcis-28493 19 13 an an DET fcis-28493 19 14 object object NOUN fcis-28493 19 15 's 's PART fcis-28493 19 16 orientation orientation NOUN fcis-28493 19 17 and and CCONJ fcis-28493 19 18 position position NOUN fcis-28493 19 19 within within ADP fcis-28493 19 20 threedimensional threedimensional ADJ fcis-28493 19 21 space space NOUN fcis-28493 19 22 , , PUNCT fcis-28493 19 23 thereby thereby ADV fcis-28493 19 24 providing provide VERB fcis-28493 19 25 more more ADV fcis-28493 19 26 accurate accurate ADJ fcis-28493 19 27 predictions prediction NOUN fcis-28493 19 28 of of ADP fcis-28493 19 29 grasping grasp VERB fcis-28493 19 30 poses pose NOUN fcis-28493 19 31 . . PUNCT fcis-28493 20 1 due due ADP fcis-28493 20 2 to to ADP fcis-28493 20 3 the the DET fcis-28493 20 4 significant significant ADJ fcis-28493 20 5 advantages advantage NOUN fcis-28493 20 6 of of ADP fcis-28493 20 7 the the DET fcis-28493 20 8 se(3 se(3 NOUN fcis-28493 20 9 ) ) PUNCT fcis-28493 20 10 method method NOUN fcis-28493 20 11 , , PUNCT fcis-28493 20 12 this this DET fcis-28493 20 13 paper paper NOUN fcis-28493 20 14 presents present VERB fcis-28493 20 15 an an DET fcis-28493 20 16 improved improved ADJ fcis-28493 20 17 design design NOUN fcis-28493 20 18 based base VERB fcis-28493 20 19 on on ADP fcis-28493 20 20 se(3 se(3 NOUN fcis-28493 20 21 ) ) PUNCT fcis-28493 20 22 , , PUNCT fcis-28493 20 23 resulting result VERB fcis-28493 20 24 in in ADP fcis-28493 20 25 a a DET fcis-28493 20 26 point point NOUN fcis-28493 20 27 cloud cloud NOUN fcis-28493 20 28 - - PUNCT fcis-28493 20 29 based base VERB fcis-28493 20 30 grasp grasp NOUN fcis-28493 20 31 detection detection NOUN fcis-28493 20 32 network network NOUN fcis-28493 20 33 structure structure NOUN fcis-28493 20 34 referred refer VERB fcis-28493 20 35 to to ADP fcis-28493 20 36 as as ADP fcis-28493 20 37 tes tes NOUN fcis-28493 20 38 - - PUNCT fcis-28493 20 39 net net ADJ fcis-28493 20 40 ( ( PUNCT fcis-28493 20 41 transformedgesage transformedgesage NOUN fcis-28493 20 42 network network NOUN fcis-28493 20 43 ) ) PUNCT fcis-28493 20 44 . . PUNCT fcis-28493 21 1 this this DET fcis-28493 21 2 network network NOUN fcis-28493 21 3 represents represent VERB fcis-28493 21 4 unique unique ADJ fcis-28493 21 5 six six NUM fcis-28493 21 6 - - PUNCT fcis-28493 21 7 degree degree NOUN fcis-28493 21 8 - - PUNCT fcis-28493 21 9 of of ADP fcis-28493 21 10 - - PUNCT fcis-28493 21 11 freedom freedom NOUN fcis-28493 21 12 grasp grasp NOUN fcis-28493 21 13 poses pose VERB fcis-28493 21 14 through through ADP fcis-28493 21 15 a a DET fcis-28493 21 16 pair pair NOUN fcis-28493 21 17 of of ADP fcis-28493 21 18 vertices vertex NOUN fcis-28493 21 19 within within ADP fcis-28493 21 20 a a DET fcis-28493 21 21 graph graph NOUN fcis-28493 21 22 . . PUNCT fcis-28493 22 1 to to PART fcis-28493 22 2 enhance enhance VERB fcis-28493 22 3 the the DET fcis-28493 22 4 model model NOUN fcis-28493 22 5 's 's PART fcis-28493 22 6 accuracy accuracy NOUN fcis-28493 22 7 and and CCONJ fcis-28493 22 8 robustness robustness NOUN fcis-28493 22 9 , , PUNCT fcis-28493 22 10 tes tes NOUN fcis-28493 22 11 - - ADJ fcis-28493 22 12 net net NOUN fcis-28493 22 13 constructs construct VERB fcis-28493 22 14 a a DET fcis-28493 22 15 knearest knearest NOUN fcis-28493 22 16 neighbor neighbor NOUN fcis-28493 22 17 graph graph NOUN fcis-28493 22 18 ( ( PUNCT fcis-28493 22 19 knn knn PROPN fcis-28493 22 20 ) ) PUNCT fcis-28493 22 21 and and CCONJ fcis-28493 22 22 selects select VERB fcis-28493 22 23 the the DET fcis-28493 22 24 top top ADJ fcis-28493 22 25 k k PROPN fcis-28493 22 26 point point NOUN fcis-28493 22 27 clouds cloud NOUN fcis-28493 22 28 for for ADP fcis-28493 22 29 building build VERB fcis-28493 22 30 an an DET fcis-28493 22 31 adjacency adjacency NOUN fcis-28493 22 32 matrix matrix NOUN fcis-28493 22 33 by by ADP fcis-28493 22 34 calculating calculate VERB fcis-28493 22 35 cosine cosine NOUN fcis-28493 22 36 distances distance NOUN fcis-28493 22 37 . . PUNCT fcis-28493 23 1 this this DET fcis-28493 23 2 approach approach NOUN fcis-28493 23 3 ensures ensure VERB fcis-28493 23 4 that that SCONJ fcis-28493 23 5 the the DET fcis-28493 23 6 network network NOUN fcis-28493 23 7 effectively effectively ADV fcis-28493 23 8 captures capture VERB fcis-28493 23 9 spatial spatial ADJ fcis-28493 23 10 relationships relationship NOUN fcis-28493 23 11 among among ADP fcis-28493 23 12 point point NOUN fcis-28493 23 13 clouds cloud NOUN fcis-28493 23 14 . . PUNCT fcis-28493 24 1 subsequently subsequently ADV fcis-28493 24 2 , , PUNCT fcis-28493 24 3 pointtransformerconv pointtransformerconv PROPN fcis-28493 24 4 is be AUX fcis-28493 24 5 employed employ VERB fcis-28493 24 6 for for ADP fcis-28493 24 7 feature feature NOUN fcis-28493 24 8 extraction extraction NOUN fcis-28493 24 9 , , PUNCT fcis-28493 24 10 while while SCONJ fcis-28493 24 11 the the DET fcis-28493 24 12 extracted extract VERB fcis-28493 24 13 point point NOUN fcis-28493 24 14 cloud cloud NOUN fcis-28493 24 15 features feature NOUN fcis-28493 24 16 are be AUX fcis-28493 24 17 sampled sample VERB fcis-28493 24 18 and and CCONJ fcis-28493 24 19 aggregated aggregate VERB fcis-28493 24 20 using use VERB fcis-28493 24 21 the the DET fcis-28493 24 22 graph graph NOUN fcis-28493 24 23 neural neural ADJ fcis-28493 24 24 module module NOUN fcis-28493 24 25 sageconv sageconv NOUN fcis-28493 24 26 to to PART fcis-28493 24 27 enrich enrich VERB fcis-28493 24 28 and and CCONJ fcis-28493 24 29 enhance enhance VERB fcis-28493 24 30 these these DET fcis-28493 24 31 features feature NOUN fcis-28493 24 32 . . PUNCT fcis-28493 25 1 finally finally ADV fcis-28493 25 2 , , PUNCT fcis-28493 25 3 by by ADP fcis-28493 25 4 thoroughly thoroughly ADV fcis-28493 25 5 mining mine VERB fcis-28493 25 6 potential potential ADJ fcis-28493 25 7 information information NOUN fcis-28493 25 8 within within ADP fcis-28493 25 9 the the DET fcis-28493 25 10 point point NOUN fcis-28493 25 11 cloud cloud NOUN fcis-28493 25 12 data datum NOUN fcis-28493 25 13 , , PUNCT fcis-28493 25 14 we we PRON fcis-28493 25 15 aim aim VERB fcis-28493 25 16 to to PART fcis-28493 25 17 improve improve VERB fcis-28493 25 18 predictions prediction NOUN fcis-28493 25 19 of of ADP fcis-28493 25 20 grasping grasp VERB fcis-28493 25 21 poses pose NOUN fcis-28493 25 22 more more ADV fcis-28493 25 23 effectively effectively ADV fcis-28493 25 24 . . PUNCT fcis-28493 26 1 2 2 X fcis-28493 26 2 . . X fcis-28493 26 3 research research NOUN fcis-28493 26 4 on on ADP fcis-28493 26 5 robotic robotic ADJ fcis-28493 26 6 arm arm NOUN fcis-28493 26 7 grasping grasp VERB fcis-28493 26 8 2.1 2.1 NUM fcis-28493 26 9 . . PUNCT fcis-28493 27 1 6 6 NUM fcis-28493 27 2 - - PUNCT fcis-28493 27 3 dof dof NOUN fcis-28493 27 4 grasping grasp VERB fcis-28493 27 5 methods method NOUN fcis-28493 27 6 six six NUM fcis-28493 27 7 degrees degree NOUN fcis-28493 27 8 of of ADP fcis-28493 27 9 freedom freedom NOUN fcis-28493 27 10 ( ( PUNCT fcis-28493 27 11 6 6 NUM fcis-28493 27 12 - - PUNCT fcis-28493 27 13 dof dof NOUN fcis-28493 27 14 ) ) PUNCT fcis-28493 27 15 grasping grasp VERB fcis-28493 27 16 methods method NOUN fcis-28493 27 17 are be AUX fcis-28493 27 18 generally generally ADV fcis-28493 27 19 categorized categorize VERB fcis-28493 27 20 into into ADP fcis-28493 27 21 two two NUM fcis-28493 27 22 main main ADJ fcis-28493 27 23 approaches[2 approaches[2 PROPN fcis-28493 27 24 ] ] PUNCT fcis-28493 27 25 . . PUNCT fcis-28493 28 1 the the DET fcis-28493 28 2 first first ADJ fcis-28493 28 3 category category NOUN fcis-28493 28 4 consists consist VERB fcis-28493 28 5 of of ADP fcis-28493 28 6 sample sample NOUN fcis-28493 28 7 - - PUNCT fcis-28493 28 8 based base VERB fcis-28493 28 9 methods method NOUN fcis-28493 28 10 , , PUNCT fcis-28493 28 11 such such ADJ fcis-28493 28 12 as as ADP fcis-28493 28 13 gpd gpd PROPN fcis-28493 28 14 , , PUNCT fcis-28493 28 15 pointnetgpd pointnetgpd ADJ fcis-28493 28 16 , , PUNCT fcis-28493 28 17 and and CCONJ fcis-28493 28 18 graspnet graspnet NOUN fcis-28493 28 19 . . PUNCT fcis-28493 29 1 these these DET fcis-28493 29 2 methods method NOUN fcis-28493 29 3 typically typically ADV fcis-28493 29 4 utilize utilize VERB fcis-28493 29 5 grasp grasp NOUN fcis-28493 29 6 sampling sampling NOUN fcis-28493 29 7 and and CCONJ fcis-28493 29 8 evaluation evaluation NOUN fcis-28493 29 9 modules module NOUN fcis-28493 29 10 to to PART fcis-28493 29 11 individually individually ADV fcis-28493 29 12 represent represent VERB fcis-28493 29 13 and and CCONJ fcis-28493 29 14 assess assess VERB fcis-28493 29 15 each each DET fcis-28493 29 16 grasp grasp NOUN fcis-28493 29 17 pose[3 pose[3 X fcis-28493 29 18 ] ] X fcis-28493 29 19 . . PUNCT fcis-28493 30 1 consequently consequently ADV fcis-28493 30 2 , , PUNCT fcis-28493 30 3 these these DET fcis-28493 30 4 approaches approach NOUN fcis-28493 30 5 tend tend VERB fcis-28493 30 6 to to PART fcis-28493 30 7 require require VERB fcis-28493 30 8 significant significant ADJ fcis-28493 30 9 computational computational ADJ fcis-28493 30 10 time time NOUN fcis-28493 30 11 during during ADP fcis-28493 30 12 both both DET fcis-28493 30 13 training training NOUN fcis-28493 30 14 and and CCONJ fcis-28493 30 15 execution execution NOUN fcis-28493 30 16 . . PUNCT fcis-28493 31 1 to to PART fcis-28493 31 2 improve improve VERB fcis-28493 31 3 computational computational ADJ fcis-28493 31 4 efficiency efficiency NOUN fcis-28493 31 5 , , PUNCT fcis-28493 31 6 the the DET fcis-28493 31 7 present present ADJ fcis-28493 31 8 study study NOUN fcis-28493 31 9 adopts adopt VERB fcis-28493 31 10 a a DET fcis-28493 31 11 strategy strategy NOUN fcis-28493 31 12 of of ADP fcis-28493 31 13 representing represent VERB fcis-28493 31 14 different different ADJ fcis-28493 31 15 grasp grasp NOUN fcis-28493 31 16 poses pose NOUN fcis-28493 31 17 through through ADP fcis-28493 31 18 shared share VERB fcis-28493 31 19 features feature NOUN fcis-28493 31 20 , , PUNCT fcis-28493 31 21 thereby thereby ADV fcis-28493 31 22 significantly significantly ADV fcis-28493 31 23 enhancing enhance VERB fcis-28493 31 24 processing processing NOUN fcis-28493 31 25 speed speed NOUN fcis-28493 31 26 and and CCONJ fcis-28493 31 27 efficiency efficiency NOUN fcis-28493 31 28 . . PUNCT fcis-28493 32 1 the the DET fcis-28493 32 2 second second ADJ fcis-28493 32 3 category category NOUN fcis-28493 32 4 encompasses encompass VERB fcis-28493 32 5 element element ADJ fcis-28493 32 6 predictionbased predictionbase VERB fcis-28493 32 7 methods method NOUN fcis-28493 32 8 , , PUNCT fcis-28493 32 9 which which PRON fcis-28493 32 10 include include VERB fcis-28493 32 11 both both DET fcis-28493 32 12 point point NOUN fcis-28493 32 13 - - PUNCT fcis-28493 32 14 based base VERB fcis-28493 32 15 and and CCONJ fcis-28493 32 16 voxelbased voxelbased ADJ fcis-28493 32 17 approaches approach NOUN fcis-28493 32 18 . . PUNCT fcis-28493 33 1 point point NOUN fcis-28493 33 2 - - PUNCT fcis-28493 33 3 based base VERB fcis-28493 33 4 methods method NOUN fcis-28493 33 5 estimate estimate VERB fcis-28493 33 6 the the DET fcis-28493 33 7 grasp grasp NOUN fcis-28493 33 8 quality quality NOUN fcis-28493 33 9 for for ADP fcis-28493 33 10 all all DET fcis-28493 33 11 relevant relevant ADJ fcis-28493 33 12 points point NOUN fcis-28493 33 13 or or CCONJ fcis-28493 33 14 voxels voxel NOUN fcis-28493 33 15 via via ADP fcis-28493 33 16 a a DET fcis-28493 33 17 single single ADJ fcis-28493 33 18 forward forward ADV fcis-28493 33 19 pass pass NOUN fcis-28493 33 20 . . PUNCT fcis-28493 34 1 for for ADP fcis-28493 34 2 example example NOUN fcis-28493 34 3 , , PUNCT fcis-28493 34 4 s4 s4 PROPN fcis-28493 34 5 g g NOUN fcis-28493 34 6 employs employ NOUN fcis-28493 34 7 pointnet++ pointnet++ VERB fcis-28493 34 8 to to PART fcis-28493 34 9 predict predict VERB fcis-28493 34 10 the the DET fcis-28493 34 11 features feature NOUN fcis-28493 34 12 and and CCONJ fcis-28493 34 13 grasp grasp VERB fcis-28493 34 14 quality quality NOUN fcis-28493 34 15 for for ADP fcis-28493 34 16 each each DET fcis-28493 34 17 individual individual ADJ fcis-28493 34 18 point[5 point[5 NOUN fcis-28493 34 19 ] ] PUNCT fcis-28493 34 20 , , PUNCT fcis-28493 34 21 while while SCONJ fcis-28493 34 22 regnet regnet NOUN fcis-28493 34 23 predicts predict VERB fcis-28493 34 24 the the DET fcis-28493 34 25 grasp grasp NOUN fcis-28493 34 26 pose pose VERB fcis-28493 34 27 by by ADP fcis-28493 34 28 regressing regress VERB fcis-28493 34 29 the the DET fcis-28493 34 30 geometry geometry NOUN fcis-28493 34 31 surrounding surround VERB fcis-28493 34 32 sampled sample VERB fcis-28493 34 33 points[7 points[7 NOUN fcis-28493 34 34 ] ] PUNCT fcis-28493 34 35 . . PUNCT fcis-28493 35 1 although although SCONJ fcis-28493 35 2 classification classification NOUN fcis-28493 35 3 - - PUNCT fcis-28493 35 4 based base VERB fcis-28493 35 5 methods method NOUN fcis-28493 35 6 are be AUX fcis-28493 35 7 capable capable ADJ fcis-28493 35 8 of of ADP fcis-28493 35 9 addressing address VERB fcis-28493 35 10 multiple multiple ADJ fcis-28493 35 11 grasp grasp NOUN fcis-28493 35 12 poses pose NOUN fcis-28493 35 13 , , PUNCT fcis-28493 35 14 they they PRON fcis-28493 35 15 generally generally ADV fcis-28493 35 16 require require VERB fcis-28493 35 17 a a DET fcis-28493 35 18 larger large ADJ fcis-28493 35 19 quantity quantity NOUN fcis-28493 35 20 of of ADP fcis-28493 35 21 training training NOUN fcis-28493 35 22 data datum NOUN fcis-28493 35 23 . . PUNCT fcis-28493 36 1 voxel voxel PROPN fcis-28493 36 2 - - PUNCT fcis-28493 36 3 based base VERB fcis-28493 36 4 methods method NOUN fcis-28493 36 5 , , PUNCT fcis-28493 36 6 on on ADP fcis-28493 36 7 the the DET fcis-28493 36 8 other other ADJ fcis-28493 36 9 hand hand NOUN fcis-28493 36 10 , , PUNCT fcis-28493 36 11 represent represent VERB fcis-28493 36 12 spatial spatial ADJ fcis-28493 36 13 information information NOUN fcis-28493 36 14 through through ADP fcis-28493 36 15 voxel voxel PROPN fcis-28493 36 16 grids grid NOUN fcis-28493 36 17 . . PUNCT fcis-28493 37 1 however however ADV fcis-28493 37 2 , , PUNCT fcis-28493 37 3 the the DET fcis-28493 37 4 memory memory NOUN fcis-28493 37 5 66 66 NUM fcis-28493 37 6 requirements requirement NOUN fcis-28493 37 7 for for ADP fcis-28493 37 8 these these DET fcis-28493 37 9 methods method NOUN fcis-28493 37 10 increase increase VERB fcis-28493 37 11 cubically cubically ADV fcis-28493 37 12 as as SCONJ fcis-28493 37 13 the the DET fcis-28493 37 14 grid grid NOUN fcis-28493 37 15 resolution resolution NOUN fcis-28493 37 16 is be AUX fcis-28493 37 17 enhanced enhance VERB fcis-28493 37 18 , , PUNCT fcis-28493 37 19 imposing impose VERB fcis-28493 37 20 significant significant ADJ fcis-28493 37 21 constraints constraint NOUN fcis-28493 37 22 on on ADP fcis-28493 37 23 their their PRON fcis-28493 37 24 applicability applicability NOUN fcis-28493 37 25 in in ADP fcis-28493 37 26 high high ADJ fcis-28493 37 27 - - PUNCT fcis-28493 37 28 resolution resolution NOUN fcis-28493 37 29 scenarios scenario NOUN fcis-28493 37 30 . . PUNCT fcis-28493 38 1 as as ADP fcis-28493 38 2 a a DET fcis-28493 38 3 result result NOUN fcis-28493 38 4 , , PUNCT fcis-28493 38 5 while while SCONJ fcis-28493 38 6 voxel voxel PROPN fcis-28493 38 7 - - PUNCT fcis-28493 38 8 based base VERB fcis-28493 38 9 methods method NOUN fcis-28493 38 10 provide provide VERB fcis-28493 38 11 structured structured ADJ fcis-28493 38 12 spatial spatial ADJ fcis-28493 38 13 representations representation NOUN fcis-28493 38 14 , , PUNCT fcis-28493 38 15 they they PRON fcis-28493 38 16 face face VERB fcis-28493 38 17 challenges challenge NOUN fcis-28493 38 18 related relate VERB fcis-28493 38 19 to to ADP fcis-28493 38 20 excessive excessive ADJ fcis-28493 38 21 memory memory NOUN fcis-28493 38 22 consumption consumption NOUN fcis-28493 38 23 and and CCONJ fcis-28493 38 24 reduced reduce VERB fcis-28493 38 25 computational computational ADJ fcis-28493 38 26 efficiency efficiency NOUN fcis-28493 38 27 when when SCONJ fcis-28493 38 28 processing process VERB fcis-28493 38 29 high high ADJ fcis-28493 38 30 - - PUNCT fcis-28493 38 31 resolution resolution NOUN fcis-28493 38 32 data datum NOUN fcis-28493 38 33 , , PUNCT fcis-28493 38 34 which which PRON fcis-28493 38 35 in in ADP fcis-28493 38 36 turn turn NOUN fcis-28493 38 37 directly directly ADV fcis-28493 38 38 affects affect VERB fcis-28493 38 39 their their PRON fcis-28493 38 40 real real ADJ fcis-28493 38 41 - - PUNCT fcis-28493 38 42 time time NOUN fcis-28493 38 43 processing processing NOUN fcis-28493 38 44 capabilities capability NOUN fcis-28493 38 45 . . PUNCT fcis-28493 39 1 2.2 2.2 NUM fcis-28493 39 2 . . PUNCT fcis-28493 39 3 grasp grasp NOUN fcis-28493 39 4 pose pose NOUN fcis-28493 39 5 problem problem NOUN fcis-28493 39 6 description description NOUN fcis-28493 39 7 the the DET fcis-28493 39 8 problem problem NOUN fcis-28493 39 9 of of ADP fcis-28493 39 10 3d 3d PROPN fcis-28493 39 11 grasp grasp NOUN fcis-28493 39 12 pose pose NOUN fcis-28493 39 13 detection detection NOUN fcis-28493 39 14 involves involve VERB fcis-28493 39 15 identifying identify VERB fcis-28493 39 16 a a DET fcis-28493 39 17 combination combination NOUN fcis-28493 39 18 of of ADP fcis-28493 39 19 positions position NOUN fcis-28493 39 20 and and CCONJ fcis-28493 39 21 orientations orientation NOUN fcis-28493 39 22 at at ADP fcis-28493 39 23 the the DET fcis-28493 39 24 robotic robotic ADJ fcis-28493 39 25 arm arm NOUN fcis-28493 39 26 's 's PART fcis-28493 39 27 end end NOUN fcis-28493 39 28 - - PUNCT fcis-28493 39 29 effector effector NOUN fcis-28493 39 30 that that PRON fcis-28493 39 31 can can AUX fcis-28493 39 32 successfully successfully ADV fcis-28493 39 33 grasp grasp VERB fcis-28493 39 34 the the DET fcis-28493 39 35 object object NOUN fcis-28493 39 36 , , PUNCT fcis-28493 39 37 based base VERB fcis-28493 39 38 on on ADP fcis-28493 39 39 the the DET fcis-28493 39 40 input input NOUN fcis-28493 39 41 point point NOUN fcis-28493 39 42 cloud cloud NOUN fcis-28493 39 43 data[8 data[8 PROPN fcis-28493 39 44 ] ] PUNCT fcis-28493 39 45 . . PUNCT fcis-28493 40 1 the the DET fcis-28493 40 2 point point NOUN fcis-28493 40 3 cloud cloud NOUN fcis-28493 40 4 is be AUX fcis-28493 40 5 represented represent VERB fcis-28493 40 6 as as SCONJ fcis-28493 40 7 shown show VERB fcis-28493 40 8 in in ADP fcis-28493 40 9 equation equation NOUN fcis-28493 40 10 ( ( PUNCT fcis-28493 40 11 1 1 NUM fcis-28493 40 12 ) ) PUNCT fcis-28493 40 13 , , PUNCT fcis-28493 40 14 where where SCONJ fcis-28493 40 15 is be AUX fcis-28493 40 16 a a DET fcis-28493 40 17 point point NOUN fcis-28493 40 18 in in ADP fcis-28493 40 19 three three NUM fcis-28493 40 20 - - PUNCT fcis-28493 40 21 dimensional dimensional ADJ fcis-28493 40 22 space space NOUN fcis-28493 40 23 , , PUNCT fcis-28493 40 24 is be AUX fcis-28493 40 25 the the DET fcis-28493 40 26 number number NOUN fcis-28493 40 27 of of ADP fcis-28493 40 28 points point NOUN fcis-28493 40 29 in in ADP fcis-28493 40 30 the the DET fcis-28493 40 31 point point NOUN fcis-28493 40 32 cloud cloud NOUN fcis-28493 40 33 , , PUNCT fcis-28493 40 34 and and CCONJ fcis-28493 40 35 represents represent VERB fcis-28493 40 36 the the DET fcis-28493 40 37 three three NUM fcis-28493 40 38 - - PUNCT fcis-28493 40 39 dimensional dimensional ADJ fcis-28493 40 40 euclidean euclidean ADJ fcis-28493 40 41 space space NOUN fcis-28493 40 42 . . PUNCT fcis-28493 41 1 ( ( PUNCT fcis-28493 41 2 1 1 X fcis-28493 41 3 ) ) PUNCT fcis-28493 41 4 the the DET fcis-28493 41 5 pose pose NOUN fcis-28493 41 6 of of ADP fcis-28493 41 7 the the DET fcis-28493 41 8 robotic robotic ADJ fcis-28493 41 9 gripper gripper NOUN fcis-28493 41 10 can can AUX fcis-28493 41 11 be be AUX fcis-28493 41 12 represented represent VERB fcis-28493 41 13 as as ADP fcis-28493 41 14 , , PUNCT fcis-28493 41 15 where where SCONJ fcis-28493 41 16 denotes denote VERB fcis-28493 41 17 the the DET fcis-28493 41 18 position position NOUN fcis-28493 41 19 of of ADP fcis-28493 41 20 the the DET fcis-28493 41 21 gripper gripper NOUN fcis-28493 41 22 's 's PART fcis-28493 41 23 center center NOUN fcis-28493 41 24 in in ADP fcis-28493 41 25 space space NOUN fcis-28493 41 26 , , PUNCT fcis-28493 41 27 and and CCONJ fcis-28493 41 28 represents represent VERB fcis-28493 41 29 its its PRON fcis-28493 41 30 orientation orientation NOUN fcis-28493 41 31 . . PUNCT fcis-28493 42 1 the the DET fcis-28493 42 2 representation representation NOUN fcis-28493 42 3 of of ADP fcis-28493 42 4 the the DET fcis-28493 42 5 grasp grasp NOUN fcis-28493 42 6 pose pose NOUN fcis-28493 42 7 is be AUX fcis-28493 42 8 shown show VERB fcis-28493 42 9 in in ADP fcis-28493 42 10 equation equation NOUN fcis-28493 42 11 ( ( PUNCT fcis-28493 42 12 2 2 NUM fcis-28493 42 13 ) ) PUNCT fcis-28493 42 14 . . PUNCT fcis-28493 43 1 ( ( PUNCT fcis-28493 43 2 2 2 X fcis-28493 43 3 ) ) PUNCT fcis-28493 43 4 the the DET fcis-28493 43 5 key key ADJ fcis-28493 43 6 challenge challenge NOUN fcis-28493 43 7 in in ADP fcis-28493 43 8 object object NOUN fcis-28493 43 9 grasping grasp VERB fcis-28493 43 10 is be AUX fcis-28493 43 11 to to PART fcis-28493 43 12 map map VERB fcis-28493 43 13 the the DET fcis-28493 43 14 point point NOUN fcis-28493 43 15 cloud cloud NOUN fcis-28493 43 16 to to ADP fcis-28493 43 17 the the DET fcis-28493 43 18 grasp grasp NOUN fcis-28493 43 19 poses pose NOUN fcis-28493 43 20 detected detect VERB fcis-28493 43 21 within within ADP fcis-28493 43 22 the the DET fcis-28493 43 23 scene scene NOUN fcis-28493 43 24 . . PUNCT fcis-28493 44 1 the the DET fcis-28493 44 2 grasp grasp NOUN fcis-28493 44 3 quality quality NOUN fcis-28493 44 4 is be AUX fcis-28493 44 5 represented represent VERB fcis-28493 44 6 by by ADP fcis-28493 44 7 a a DET fcis-28493 44 8 function function NOUN fcis-28493 44 9 that that PRON fcis-28493 44 10 indicates indicate VERB fcis-28493 44 11 the the DET fcis-28493 44 12 quality quality NOUN fcis-28493 44 13 of of ADP fcis-28493 44 14 the the DET fcis-28493 44 15 grasp grasp NOUN fcis-28493 44 16 pose pose NOUN fcis-28493 44 17 . . PUNCT fcis-28493 45 1 when when SCONJ fcis-28493 45 2 the the DET fcis-28493 45 3 same same ADJ fcis-28493 45 4 rotation rotation NOUN fcis-28493 45 5 and and CCONJ fcis-28493 45 6 translation translation NOUN fcis-28493 45 7 transformations transformation NOUN fcis-28493 45 8 are be AUX fcis-28493 45 9 applied apply VERB fcis-28493 45 10 to to ADP fcis-28493 45 11 both both CCONJ fcis-28493 45 12 the the DET fcis-28493 45 13 object object NOUN fcis-28493 45 14 and and CCONJ fcis-28493 45 15 the the DET fcis-28493 45 16 grasp grasp NOUN fcis-28493 45 17 pose pose VERB fcis-28493 45 18 , , PUNCT fcis-28493 45 19 the the DET fcis-28493 45 20 predicted predict VERB fcis-28493 45 21 grasp grasp NOUN fcis-28493 45 22 quality quality NOUN fcis-28493 45 23 should should AUX fcis-28493 45 24 remain remain VERB fcis-28493 45 25 invariant invariant ADJ fcis-28493 45 26 . . PUNCT fcis-28493 46 1 therefore therefore ADV fcis-28493 46 2 , , PUNCT fcis-28493 46 3 this this DET fcis-28493 46 4 paper paper NOUN fcis-28493 46 5 applies apply VERB fcis-28493 46 6 centering center VERB fcis-28493 46 7 and and CCONJ fcis-28493 46 8 random random ADJ fcis-28493 46 9 rotation rotation NOUN fcis-28493 46 10 augmentation augmentation NOUN fcis-28493 46 11 to to ADP fcis-28493 46 12 the the DET fcis-28493 46 13 point point NOUN fcis-28493 46 14 cloud cloud NOUN fcis-28493 46 15 data datum NOUN fcis-28493 46 16 to to PART fcis-28493 46 17 improve improve VERB fcis-28493 46 18 data datum NOUN fcis-28493 46 19 quality quality NOUN fcis-28493 46 20 and and CCONJ fcis-28493 46 21 enhance enhance VERB fcis-28493 46 22 the the DET fcis-28493 46 23 model model NOUN fcis-28493 46 24 's 's PART fcis-28493 46 25 generalization generalization NOUN fcis-28493 46 26 capability capability NOUN fcis-28493 46 27 . . PUNCT fcis-28493 47 1 2.3 2.3 NUM fcis-28493 47 2 . . PUNCT fcis-28493 48 1 definition definition NOUN fcis-28493 48 2 of of ADP fcis-28493 48 3 robotic robotic ADJ fcis-28493 48 4 arm arm NOUN fcis-28493 48 5 grasp grasp NOUN fcis-28493 48 6 pose pose VERB fcis-28493 48 7 in in ADP fcis-28493 48 8 this this DET fcis-28493 48 9 paper paper NOUN fcis-28493 48 10 , , PUNCT fcis-28493 48 11 a a DET fcis-28493 48 12 grasp grasp NOUN fcis-28493 48 13 is be AUX fcis-28493 48 14 represented represent VERB fcis-28493 48 15 as as ADP fcis-28493 48 16 a a DET fcis-28493 48 17 pair pair NOUN fcis-28493 48 18 of of ADP fcis-28493 48 19 points point NOUN fcis-28493 48 20 in in ADP fcis-28493 48 21 the the DET fcis-28493 48 22 point point NOUN fcis-28493 48 23 cloud cloud NOUN fcis-28493 48 24 : : PUNCT fcis-28493 48 25 the the DET fcis-28493 48 26 approach approach NOUN fcis-28493 48 27 point point NOUN fcis-28493 48 28 and and CCONJ fcis-28493 48 29 the the DET fcis-28493 48 30 contact contact NOUN fcis-28493 48 31 point point NOUN fcis-28493 48 32 . . PUNCT fcis-28493 49 1 the the DET fcis-28493 49 2 approach approach NOUN fcis-28493 49 3 point point NOUN fcis-28493 49 4 is be AUX fcis-28493 49 5 the the DET fcis-28493 49 6 position position NOUN fcis-28493 49 7 where where SCONJ fcis-28493 49 8 the the DET fcis-28493 49 9 gripper gripper NOUN fcis-28493 49 10 is be AUX fcis-28493 49 11 close close ADJ fcis-28493 49 12 to to ADP fcis-28493 49 13 the the DET fcis-28493 49 14 object object NOUN fcis-28493 49 15 , , PUNCT fcis-28493 49 16 while while SCONJ fcis-28493 49 17 the the DET fcis-28493 49 18 contact contact NOUN fcis-28493 49 19 point point NOUN fcis-28493 49 20 is be AUX fcis-28493 49 21 where where SCONJ fcis-28493 49 22 the the DET fcis-28493 49 23 gripper gripper NOUN fcis-28493 49 24 makes make VERB fcis-28493 49 25 contact contact NOUN fcis-28493 49 26 with with ADP fcis-28493 49 27 the the DET fcis-28493 49 28 object object NOUN fcis-28493 49 29 . . PUNCT fcis-28493 50 1 the the DET fcis-28493 50 2 grasp grasp NOUN fcis-28493 50 3 direction direction NOUN fcis-28493 50 4 is be AUX fcis-28493 50 5 defined define VERB fcis-28493 50 6 by by ADP fcis-28493 50 7 estimating estimate VERB fcis-28493 50 8 the the DET fcis-28493 50 9 normal normal ADJ fcis-28493 50 10 at at ADP fcis-28493 50 11 the the DET fcis-28493 50 12 approach approach NOUN fcis-28493 50 13 point point NOUN fcis-28493 50 14 . . PUNCT fcis-28493 51 1 the the DET fcis-28493 51 2 gripper gripper NOUN fcis-28493 51 3 's 's PART fcis-28493 51 4 end end NOUN fcis-28493 51 5 - - PUNCT fcis-28493 51 6 effector effector NOUN fcis-28493 51 7 moves move NOUN fcis-28493 51 8 parallel parallel ADJ fcis-28493 51 9 to to ADP fcis-28493 51 10 the the DET fcis-28493 51 11 vector vector NOUN fcis-28493 51 12 and and CCONJ fcis-28493 51 13 approaches approach VERB fcis-28493 51 14 the the DET fcis-28493 51 15 object object NOUN fcis-28493 51 16 along along ADP fcis-28493 51 17 the the DET fcis-28493 51 18 vector vector NOUN fcis-28493 51 19 . . PUNCT fcis-28493 52 1 the the DET fcis-28493 52 2 gripper gripper NOUN fcis-28493 52 3 's 's PART fcis-28493 52 4 center center NOUN fcis-28493 52 5 is be AUX fcis-28493 52 6 positioned position VERB fcis-28493 52 7 at at ADP fcis-28493 52 8 the the DET fcis-28493 52 9 ideal ideal ADJ fcis-28493 52 10 contact contact NOUN fcis-28493 52 11 point point NOUN fcis-28493 52 12 at at ADP fcis-28493 52 13 the the DET fcis-28493 52 14 midpoint midpoint NOUN fcis-28493 52 15 of of ADP fcis-28493 52 16 the the DET fcis-28493 52 17 fingers finger NOUN fcis-28493 52 18 . . PUNCT fcis-28493 53 1 fig fig NOUN fcis-28493 53 2 1 1 NUM fcis-28493 53 3 . . PUNCT fcis-28493 53 4 diagram diagram NOUN fcis-28493 53 5 of of ADP fcis-28493 53 6 grasp grasp NOUN fcis-28493 53 7 pose pose NOUN fcis-28493 53 8 representation representation NOUN fcis-28493 53 9 3 3 NUM fcis-28493 53 10 . . PUNCT fcis-28493 54 1 network network NOUN fcis-28493 54 2 model model NOUN fcis-28493 54 3 3.1 3.1 NUM fcis-28493 54 4 . . PUNCT fcis-28493 55 1 structure structure NOUN fcis-28493 55 2 design design NOUN fcis-28493 55 3 of of ADP fcis-28493 55 4 the the DET fcis-28493 55 5 grasp grasp NOUN fcis-28493 55 6 detection detection NOUN fcis-28493 55 7 model model NOUN fcis-28493 55 8 feature feature NOUN fcis-28493 55 9 extraction extraction NOUN fcis-28493 55 10 module module NOUN fcis-28493 55 11 : : PUNCT fcis-28493 55 12 this this DET fcis-28493 55 13 module module NOUN fcis-28493 55 14 extracts extract NOUN fcis-28493 55 15 useful useful ADJ fcis-28493 55 16 geometric geometric ADJ fcis-28493 55 17 features feature NOUN fcis-28493 55 18 from from ADP fcis-28493 55 19 the the DET fcis-28493 55 20 input input NOUN fcis-28493 55 21 point point NOUN fcis-28493 55 22 cloud cloud NOUN fcis-28493 55 23 data datum NOUN fcis-28493 55 24 . . PUNCT fcis-28493 56 1 by by ADP fcis-28493 56 2 using use VERB fcis-28493 56 3 methods method NOUN fcis-28493 56 4 such such ADJ fcis-28493 56 5 as as ADP fcis-28493 56 6 k k NOUN fcis-28493 56 7 - - PUNCT fcis-28493 56 8 nearest near ADJ fcis-28493 56 9 neighbors neighbor NOUN fcis-28493 56 10 ( ( PUNCT fcis-28493 56 11 knn knn PROPN fcis-28493 56 12 ) ) PUNCT fcis-28493 56 13 , , PUNCT fcis-28493 56 14 it it PRON fcis-28493 56 15 constructs construct VERB fcis-28493 56 16 neighborhood neighborhood NOUN fcis-28493 56 17 relationships relationship NOUN fcis-28493 56 18 between between ADP fcis-28493 56 19 points point NOUN fcis-28493 56 20 and and CCONJ fcis-28493 56 21 aggregates aggregate NOUN fcis-28493 56 22 local local ADJ fcis-28493 56 23 features feature NOUN fcis-28493 56 24 through through ADP fcis-28493 56 25 specific specific ADJ fcis-28493 56 26 convolution convolution NOUN fcis-28493 56 27 operations operation NOUN fcis-28493 56 28 , , PUNCT fcis-28493 56 29 thus thus ADV fcis-28493 56 30 generating generate VERB fcis-28493 56 31 a a DET fcis-28493 56 32 local local ADJ fcis-28493 56 33 feature feature NOUN fcis-28493 56 34 description description NOUN fcis-28493 56 35 for for ADP fcis-28493 56 36 each each DET fcis-28493 56 37 point point NOUN fcis-28493 56 38 . . PUNCT fcis-28493 57 1 graph graph NOUN fcis-28493 57 2 neural neural ADJ fcis-28493 57 3 network network NOUN fcis-28493 57 4 module module NOUN fcis-28493 57 5 : : PUNCT fcis-28493 57 6 based base VERB fcis-28493 57 7 on on ADP fcis-28493 57 8 the the DET fcis-28493 57 9 local local ADJ fcis-28493 57 10 features feature NOUN fcis-28493 57 11 generated generate VERB fcis-28493 57 12 by by ADP fcis-28493 57 13 the the DET fcis-28493 57 14 feature feature NOUN fcis-28493 57 15 extraction extraction NOUN fcis-28493 57 16 module module NOUN fcis-28493 57 17 , , PUNCT fcis-28493 57 18 this this DET fcis-28493 57 19 module module NOUN fcis-28493 57 20 further further ADJ fcis-28493 57 21 processes process NOUN fcis-28493 57 22 these these DET fcis-28493 57 23 features feature NOUN fcis-28493 57 24 using use VERB fcis-28493 57 25 graph graph NOUN fcis-28493 57 26 convolutions convolution NOUN fcis-28493 57 27 . . PUNCT fcis-28493 58 1 graph graph NOUN fcis-28493 58 2 convolutions convolution NOUN fcis-28493 58 3 effectively effectively ADV fcis-28493 58 4 capture capture VERB fcis-28493 58 5 both both CCONJ fcis-28493 58 6 the the DET fcis-28493 58 7 global global ADJ fcis-28493 58 8 and and CCONJ fcis-28493 58 9 local local ADJ fcis-28493 58 10 structural structural ADJ fcis-28493 58 11 information information NOUN fcis-28493 58 12 of of ADP fcis-28493 58 13 the the DET fcis-28493 58 14 point point NOUN fcis-28493 58 15 cloud cloud NOUN fcis-28493 58 16 , , PUNCT fcis-28493 58 17 enhancing enhance VERB fcis-28493 58 18 the the DET fcis-28493 58 19 expressive expressive ADJ fcis-28493 58 20 power power NOUN fcis-28493 58 21 of of ADP fcis-28493 58 22 the the DET fcis-28493 58 23 features feature NOUN fcis-28493 58 24 . . PUNCT fcis-28493 59 1 classification classification NOUN fcis-28493 59 2 and and CCONJ fcis-28493 59 3 evaluation evaluation NOUN fcis-28493 59 4 module module NOUN fcis-28493 59 5 : : PUNCT fcis-28493 59 6 this this DET fcis-28493 59 7 module module NOUN fcis-28493 59 8 performs perform VERB fcis-28493 59 9 the the DET fcis-28493 59 10 final final ADJ fcis-28493 59 11 classification classification NOUN fcis-28493 59 12 on on ADP fcis-28493 59 13 the the DET fcis-28493 59 14 processed process VERB fcis-28493 59 15 features feature NOUN fcis-28493 59 16 . . PUNCT fcis-28493 60 1 through through ADP fcis-28493 60 2 a a DET fcis-28493 60 3 multi multi ADJ fcis-28493 60 4 - - ADJ fcis-28493 60 5 layer layer ADJ fcis-28493 60 6 perceptron perceptron NOUN fcis-28493 60 7 ( ( PUNCT fcis-28493 60 8 mlp mlp NOUN fcis-28493 60 9 ) ) PUNCT fcis-28493 60 10 network network NOUN fcis-28493 60 11 , , PUNCT fcis-28493 60 12 combined combine VERB fcis-28493 60 13 with with ADP fcis-28493 60 14 the the DET fcis-28493 60 15 sigmoid sigmoid NOUN fcis-28493 60 16 activation activation NOUN fcis-28493 60 17 function function NOUN fcis-28493 60 18 , , PUNCT fcis-28493 60 19 it it PRON fcis-28493 60 20 evaluates evaluate VERB fcis-28493 60 21 each each DET fcis-28493 60 22 candidate candidate NOUN fcis-28493 60 23 grasp grasp NOUN fcis-28493 60 24 pose pose VERB fcis-28493 60 25 and and CCONJ fcis-28493 60 26 outputs output VERB fcis-28493 60 27 the the DET fcis-28493 60 28 probability probability NOUN fcis-28493 60 29 of of ADP fcis-28493 60 30 grasp grasp NOUN fcis-28493 60 31 success success NOUN fcis-28493 60 32 or or CCONJ fcis-28493 60 33 the the DET fcis-28493 60 34 classification classification NOUN fcis-28493 60 35 result result NOUN fcis-28493 60 36 , , PUNCT fcis-28493 60 37 ultimately ultimately ADV fcis-28493 60 38 determining determine VERB fcis-28493 60 39 the the DET fcis-28493 60 40 best good ADJ fcis-28493 60 41 grasp grasp NOUN fcis-28493 60 42 pose pose NOUN fcis-28493 60 43 . . PUNCT fcis-28493 61 1 object object NOUN fcis-28493 61 2     SPACE fcis-28493 61 3 point point NOUN fcis-28493 61 4     SPACE fcis-28493 61 5 cloud cloud ADJ fcis-28493 61 6 point point NOUN fcis-28493 61 7     SPACE fcis-28493 61 8 cloud cloud NOUN fcis-28493 61 9     SPACE fcis-28493 61 10 processing processing NOUN fcis-28493 61 11     SPACE fcis-28493 61 12 module module NOUN fcis-28493 61 13 feature feature NOUN fcis-28493 61 14     SPACE fcis-28493 61 15 extraction extraction NOUN fcis-28493 61 16     SPACE fcis-28493 61 17 module module NOUN fcis-28493 61 18 graph graph NOUN fcis-28493 61 19     SPACE fcis-28493 61 20 neural neural ADJ fcis-28493 61 21     SPACE fcis-28493 61 22 network network NOUN fcis-28493 61 23     SPACE fcis-28493 61 24 module module NOUN fcis-28493 61 25 feature feature NOUN fcis-28493 61 26     SPACE fcis-28493 61 27 processing processing NOUN fcis-28493 61 28     SPACE fcis-28493 61 29 module module NOUN fcis-28493 61 30 classification classification NOUN fcis-28493 61 31     SPACE fcis-28493 61 32 and and CCONJ fcis-28493 61 33     SPACE fcis-28493 61 34 evaluation evaluation NOUN fcis-28493 61 35     SPACE fcis-28493 61 36 module module NOUN fcis-28493 61 37 fig fig NOUN fcis-28493 61 38 2 2 NUM fcis-28493 61 39 . . PUNCT fcis-28493 61 40 edge edge NOUN fcis-28493 61 41 grasp grasp NOUN fcis-28493 61 42 detection detection NOUN fcis-28493 61 43 network network NOUN fcis-28493 61 44 model model NOUN fcis-28493 61 45 framework framework NOUN fcis-28493 61 46 based base VERB fcis-28493 61 47 on on ADP fcis-28493 61 48 point point NOUN fcis-28493 61 49 cloud cloud PROPN fcis-28493 61 50 3.2 3.2 NUM fcis-28493 61 51 . . PUNCT fcis-28493 62 1 grasp grasp NOUN fcis-28493 62 2 detection detection NOUN fcis-28493 62 3 network network NOUN fcis-28493 62 4 based base VERB fcis-28493 62 5 on on ADP fcis-28493 62 6 tesnet tesnet PROPN fcis-28493 62 7 p p PROPN fcis-28493 62 8 oin oin X fcis-28493 62 9 tt tt PROPN fcis-28493 62 10 ran run VERB fcis-28493 62 11 sform sform NOUN fcis-28493 62 12 ercon ercon NOUN fcis-28493 62 13 v v ADP fcis-28493 62 14 g g PROPN fcis-28493 62 15 rap rap NOUN fcis-28493 62 16 gh gh PROPN fcis-28493 62 17 sa sa PROPN fcis-28493 63 1 g g PROPN fcis-28493 63 2 e e PROPN fcis-28493 63 3 p p PROPN fcis-28493 63 4 oin oin PROPN fcis-28493 63 5 tt tt PROPN fcis-28493 63 6 ran run VERB fcis-28493 63 7 sform sform NOUN fcis-28493 63 8 ercon ercon NOUN fcis-28493 63 9 v v ADP fcis-28493 63 10 g g PROPN fcis-28493 63 11 rap rap NOUN fcis-28493 63 12 gh gh PROPN fcis-28493 63 13 sa sa PROPN fcis-28493 64 1 g g PROPN fcis-28493 64 2 e e PROPN fcis-28493 64 3 p p PROPN fcis-28493 64 4 oin oin PROPN fcis-28493 64 5 tt tt PROPN fcis-28493 64 6 ran run VERB fcis-28493 64 7 sform sform NOUN fcis-28493 64 8 ercon ercon NOUN fcis-28493 64 9 v v ADP fcis-28493 64 10 g g PROPN fcis-28493 64 11 rap rap NOUN fcis-28493 64 12 gh gh PROPN fcis-28493 64 13 sa sa PROPN fcis-28493 65 1 g g PROPN fcis-28493 65 2 e e PROPN fcis-28493 65 3 g g PROPN fcis-28493 65 4 lo lo PROPN fcis-28493 65 5 b b PROPN fcis-28493 65 6 al al PROPN fcis-28493 65 7     SPACE fcis-28493 65 8 m m PROPN fcis-28493 65 9 a a DET fcis-28493 65 10 x x NOUN fcis-28493 65 11     SPACE fcis-28493 65 12 p p NOUN fcis-28493 66 1 o o X fcis-28493 66 2 o o X fcis-28493 66 3 lin lin PROPN fcis-28493 66 4 g g PROPN fcis-28493 66 5 m m VERB fcis-28493 66 6 lp lp NOUN fcis-28493 66 7 g g PROPN fcis-28493 66 8 lo lo PROPN fcis-28493 66 9 b b PROPN fcis-28493 66 10 al al PROPN fcis-28493 66 11     SPACE fcis-28493 66 12 m m PROPN fcis-28493 66 13 a a DET fcis-28493 66 14 x x NOUN fcis-28493 66 15     SPACE fcis-28493 66 16 p p NOUN fcis-28493 67 1 o o X fcis-28493 67 2 o o X fcis-28493 67 3 lin lin PROPN fcis-28493 67 4 g g PROPN fcis-28493 67 5 m m VERB fcis-28493 67 6 lp lp PROPN fcis-28493 67 7 fig fig NOUN fcis-28493 67 8 3 3 X fcis-28493 67 9 . . PUNCT fcis-28493 67 10 grasp grasp NOUN fcis-28493 67 11 detection detection NOUN fcis-28493 67 12 network network NOUN fcis-28493 67 13 based base VERB fcis-28493 67 14 on on ADP fcis-28493 67 15 tes tes PROPN fcis-28493 67 16 - - NOUN fcis-28493 67 17 net net ADJ fcis-28493 67 18 67 67 NUM fcis-28493 67 19 this this DET fcis-28493 67 20 paper paper NOUN fcis-28493 67 21 mainly mainly ADV fcis-28493 67 22 focuses focus VERB fcis-28493 67 23 on on ADP fcis-28493 67 24 the the DET fcis-28493 67 25 edge edge NOUN fcis-28493 67 26 grasp grasp NOUN fcis-28493 67 27 detection detection NOUN fcis-28493 67 28 network network NOUN fcis-28493 67 29 ( ( PUNCT fcis-28493 67 30 edgegrasp edgegrasp NOUN fcis-28493 67 31 - - PUNCT fcis-28493 67 32 net net NOUN fcis-28493 67 33 ) ) PUNCT fcis-28493 67 34 and and CCONJ fcis-28493 67 35 proposes propose VERB fcis-28493 67 36 an an DET fcis-28493 67 37 improved improved ADJ fcis-28493 67 38 point point NOUN fcis-28493 67 39 cloud cloud NOUN fcis-28493 67 40 - - PUNCT fcis-28493 67 41 based base VERB fcis-28493 67 42 grasp grasp NOUN fcis-28493 67 43 detection detection NOUN fcis-28493 67 44 network network NOUN fcis-28493 67 45 model model NOUN fcis-28493 67 46 , , PUNCT fcis-28493 67 47 tes tes NOUN fcis-28493 67 48 - - NOUN fcis-28493 67 49 net net NOUN fcis-28493 67 50 . . PUNCT fcis-28493 68 1 this this DET fcis-28493 68 2 method method NOUN fcis-28493 68 3 introduces introduce VERB fcis-28493 68 4 pointtransformerconv pointtransformerconv ADJ fcis-28493 68 5 and and CCONJ fcis-28493 68 6 sageconv[9 sageconv[9 PRON fcis-28493 68 7 ] ] PUNCT fcis-28493 68 8 into into ADP fcis-28493 68 9 the the DET fcis-28493 68 10 edge edge NOUN fcis-28493 68 11 grasp grasp NOUN fcis-28493 68 12 network network NOUN fcis-28493 68 13 to to PART fcis-28493 68 14 enhance enhance VERB fcis-28493 68 15 its its PRON fcis-28493 68 16 ability ability NOUN fcis-28493 68 17 to to PART fcis-28493 68 18 capture capture VERB fcis-28493 68 19 global global ADJ fcis-28493 68 20 information information NOUN fcis-28493 68 21 , , PUNCT fcis-28493 68 22 thereby thereby ADV fcis-28493 68 23 improving improve VERB fcis-28493 68 24 the the DET fcis-28493 68 25 network network NOUN fcis-28493 68 26 's 's PART fcis-28493 68 27 feature feature NOUN fcis-28493 68 28 extraction extraction NOUN fcis-28493 68 29 and and CCONJ fcis-28493 68 30 expressive expressive ADJ fcis-28493 68 31 capabilities capability NOUN fcis-28493 68 32 . . PUNCT fcis-28493 69 1 the the DET fcis-28493 69 2 network network NOUN fcis-28493 69 3 framework framework NOUN fcis-28493 69 4 diagram diagram NOUN fcis-28493 69 5 is be AUX fcis-28493 69 6 shown show VERB fcis-28493 69 7 in in ADP fcis-28493 69 8 fig fig NOUN fcis-28493 69 9 . . PUNCT fcis-28493 70 1 3 3 X fcis-28493 70 2 . . X fcis-28493 70 3 as as SCONJ fcis-28493 70 4 shown show VERB fcis-28493 70 5 in in ADP fcis-28493 70 6 fig fig NOUN fcis-28493 70 7 . . PUNCT fcis-28493 71 1 3 3 NUM fcis-28493 71 2 , , PUNCT fcis-28493 71 3 tes tes NOUN fcis-28493 71 4 - - ADJ fcis-28493 71 5 net net ADJ fcis-28493 71 6 first first ADJ fcis-28493 71 7 preprocesses preprocesse NOUN fcis-28493 71 8 the the DET fcis-28493 71 9 input input NOUN fcis-28493 71 10 point point NOUN fcis-28493 71 11 cloud cloud NOUN fcis-28493 71 12 data datum NOUN fcis-28493 71 13 by by ADP fcis-28493 71 14 constructing construct VERB fcis-28493 71 15 a a DET fcis-28493 71 16 k k ADV fcis-28493 71 17 - - PUNCT fcis-28493 71 18 nearest near ADJ fcis-28493 71 19 neighbor neighbor NOUN fcis-28493 71 20 ( ( PUNCT fcis-28493 71 21 knn knn PROPN fcis-28493 71 22 ) ) PUNCT fcis-28493 71 23 graph graph NOUN fcis-28493 71 24 and and CCONJ fcis-28493 71 25 extracts extract VERB fcis-28493 71 26 local local ADJ fcis-28493 71 27 subsets subset NOUN fcis-28493 71 28 related relate VERB fcis-28493 71 29 to to ADP fcis-28493 71 30 the the DET fcis-28493 71 31 target target NOUN fcis-28493 71 32 grasp grasp NOUN fcis-28493 71 33 region region NOUN fcis-28493 71 34 . . PUNCT fcis-28493 72 1 then then ADV fcis-28493 72 2 , , PUNCT fcis-28493 72 3 features feature NOUN fcis-28493 72 4 are be AUX fcis-28493 72 5 extracted extract VERB fcis-28493 72 6 point point NOUN fcis-28493 72 7 - - PUNCT fcis-28493 72 8 by by ADP fcis-28493 72 9 - - PUNCT fcis-28493 72 10 point point NOUN fcis-28493 72 11 using use VERB fcis-28493 72 12 multiple multiple ADJ fcis-28493 72 13 pointtransformer pointtransformer NOUN fcis-28493 72 14 [ [ X fcis-28493 72 15 10]and 10]and NUM fcis-28493 72 16 graphsage graphsage NOUN fcis-28493 72 17 layers layer NOUN fcis-28493 72 18 . . PUNCT fcis-28493 73 1 the the DET fcis-28493 73 2 pointtransformer pointtransformer NOUN fcis-28493 73 3 enhances enhance VERB fcis-28493 73 4 the the DET fcis-28493 73 5 relationships relationship NOUN fcis-28493 73 6 between between ADP fcis-28493 73 7 points point NOUN fcis-28493 73 8 through through ADP fcis-28493 73 9 an an DET fcis-28493 73 10 attention attention NOUN fcis-28493 73 11 mechanism mechanism NOUN fcis-28493 73 12 , , PUNCT fcis-28493 73 13 while while SCONJ fcis-28493 73 14 graphsage graphsage NOUN fcis-28493 73 15 aggregates aggregate VERB fcis-28493 73 16 neighboring neighbor VERB fcis-28493 73 17 node node ADJ fcis-28493 73 18 information information NOUN fcis-28493 73 19 to to PART fcis-28493 73 20 further far ADV fcis-28493 73 21 enrich enrich VERB fcis-28493 73 22 the the DET fcis-28493 73 23 feature feature NOUN fcis-28493 73 24 representation representation NOUN fcis-28493 73 25 of of ADP fcis-28493 73 26 each each DET fcis-28493 73 27 point point NOUN fcis-28493 73 28 . . PUNCT fcis-28493 74 1 after after ADP fcis-28493 74 2 the the DET fcis-28493 74 3 point point NOUN fcis-28493 74 4 - - PUNCT fcis-28493 74 5 wise wise ADJ fcis-28493 74 6 feature feature NOUN fcis-28493 74 7 extraction extraction NOUN fcis-28493 74 8 , , PUNCT fcis-28493 74 9 global global ADJ fcis-28493 74 10 max max PROPN fcis-28493 74 11 pooling pooling PROPN fcis-28493 74 12 ( ( PUNCT fcis-28493 74 13 globalmaxpooling globalmaxpooling NOUN fcis-28493 74 14 ) ) PUNCT fcis-28493 74 15 is be AUX fcis-28493 74 16 applied apply VERB fcis-28493 74 17 to to PART fcis-28493 74 18 aggregate aggregate VERB fcis-28493 74 19 all all DET fcis-28493 74 20 point point NOUN fcis-28493 74 21 features feature NOUN fcis-28493 74 22 into into ADP fcis-28493 74 23 a a DET fcis-28493 74 24 global global ADJ fcis-28493 74 25 feature feature NOUN fcis-28493 74 26 vector vector NOUN fcis-28493 74 27 that that PRON fcis-28493 74 28 represents represent VERB fcis-28493 74 29 the the DET fcis-28493 74 30 entire entire ADJ fcis-28493 74 31 point point NOUN fcis-28493 74 32 cloud cloud NOUN fcis-28493 74 33 . . PUNCT fcis-28493 75 1 this this DET fcis-28493 75 2 global global ADJ fcis-28493 75 3 feature feature NOUN fcis-28493 75 4 vector vector NOUN fcis-28493 75 5 is be AUX fcis-28493 75 6 then then ADV fcis-28493 75 7 processed process VERB fcis-28493 75 8 non non ADJ fcis-28493 75 9 - - ADJ fcis-28493 75 10 linearly linearly ADV fcis-28493 75 11 through through ADP fcis-28493 75 12 a a DET fcis-28493 75 13 multilayer multilayer ADJ fcis-28493 75 14 perceptron perceptron NOUN fcis-28493 75 15 ( ( PUNCT fcis-28493 75 16 mlp mlp PROPN fcis-28493 75 17 ) ) PUNCT fcis-28493 75 18 , , PUNCT fcis-28493 75 19 serving serve VERB fcis-28493 75 20 as as ADP fcis-28493 75 21 the the DET fcis-28493 75 22 global global ADJ fcis-28493 75 23 context context PROPN fcis-28493 75 24 information information NOUN fcis-28493 75 25 for for ADP fcis-28493 75 26 subsequent subsequent ADJ fcis-28493 75 27 grasp grasp NOUN fcis-28493 75 28 prediction prediction NOUN fcis-28493 75 29 . . PUNCT fcis-28493 76 1 the the DET fcis-28493 76 2 global global ADJ fcis-28493 76 3 feature feature NOUN fcis-28493 76 4 vector vector NOUN fcis-28493 76 5 is be AUX fcis-28493 76 6 combined combine VERB fcis-28493 76 7 with with ADP fcis-28493 76 8 the the DET fcis-28493 76 9 point point NOUN fcis-28493 76 10 - - PUNCT fcis-28493 76 11 wise wise ADJ fcis-28493 76 12 features feature NOUN fcis-28493 76 13 to to PART fcis-28493 76 14 generate generate VERB fcis-28493 76 15 edge edge NOUN fcis-28493 76 16 features feature NOUN fcis-28493 76 17 for for ADP fcis-28493 76 18 each each DET fcis-28493 76 19 point point NOUN fcis-28493 76 20 , , PUNCT fcis-28493 76 21 ensuring ensure VERB fcis-28493 76 22 that that SCONJ fcis-28493 76 23 each each DET fcis-28493 76 24 point point NOUN fcis-28493 76 25 's 's PART fcis-28493 76 26 features feature NOUN fcis-28493 76 27 contain contain VERB fcis-28493 76 28 both both DET fcis-28493 76 29 local local ADJ fcis-28493 76 30 geometric geometric ADJ fcis-28493 76 31 information information NOUN fcis-28493 76 32 and and CCONJ fcis-28493 76 33 global global ADJ fcis-28493 76 34 structural structural ADJ fcis-28493 76 35 information information NOUN fcis-28493 76 36 . . PUNCT fcis-28493 77 1 finally finally ADV fcis-28493 77 2 , , PUNCT fcis-28493 77 3 the the DET fcis-28493 77 4 network network NOUN fcis-28493 77 5 uses use VERB fcis-28493 77 6 these these DET fcis-28493 77 7 edge edge NOUN fcis-28493 77 8 features feature NOUN fcis-28493 77 9 to to PART fcis-28493 77 10 predict predict VERB fcis-28493 77 11 grasp grasp NOUN fcis-28493 77 12 positions position NOUN fcis-28493 77 13 , , PUNCT fcis-28493 77 14 outputting output VERB fcis-28493 77 15 the the DET fcis-28493 77 16 best good ADJ fcis-28493 77 17 grasp grasp NOUN fcis-28493 77 18 region region NOUN fcis-28493 77 19 and and CCONJ fcis-28493 77 20 orientation orientation NOUN fcis-28493 77 21 , , PUNCT fcis-28493 77 22 thereby thereby ADV fcis-28493 77 23 providing provide VERB fcis-28493 77 24 precise precise ADJ fcis-28493 77 25 guidance guidance NOUN fcis-28493 77 26 for for ADP fcis-28493 77 27 the the DET fcis-28493 77 28 robotic robotic ADJ fcis-28493 77 29 arm arm NOUN fcis-28493 77 30 's 's PART fcis-28493 77 31 object object NOUN fcis-28493 77 32 detection detection NOUN fcis-28493 77 33 and and CCONJ fcis-28493 77 34 grasping grasp VERB fcis-28493 77 35 localization localization NOUN fcis-28493 77 36 . . PUNCT fcis-28493 78 1 3.3 3.3 NUM fcis-28493 78 2 . . PUNCT fcis-28493 79 1 pointtransformerconv pointtransformerconv ADJ fcis-28493 79 2 layer layer NOUN fcis-28493 79 3 feature feature NOUN fcis-28493 79 4 extraction extraction NOUN fcis-28493 79 5 is be AUX fcis-28493 79 6 crucial crucial ADJ fcis-28493 79 7 in in ADP fcis-28493 79 8 deep deep ADJ fcis-28493 79 9 learning learning NOUN fcis-28493 79 10 tasks task NOUN fcis-28493 79 11 for for ADP fcis-28493 79 12 point point NOUN fcis-28493 79 13 cloud cloud NOUN fcis-28493 79 14 data datum NOUN fcis-28493 79 15 . . PUNCT fcis-28493 80 1 traditional traditional ADJ fcis-28493 80 2 convolutional convolutional ADJ fcis-28493 80 3 networks network NOUN fcis-28493 80 4 rely rely VERB fcis-28493 80 5 on on ADP fcis-28493 80 6 regular regular ADJ fcis-28493 80 7 grids grid NOUN fcis-28493 80 8 , , PUNCT fcis-28493 80 9 which which PRON fcis-28493 80 10 reduces reduce VERB fcis-28493 80 11 efficiency efficiency NOUN fcis-28493 80 12 for for ADP fcis-28493 80 13 unstructured unstructured ADJ fcis-28493 80 14 point point NOUN fcis-28493 80 15 clouds cloud NOUN fcis-28493 80 16 . . PUNCT fcis-28493 81 1 to to PART fcis-28493 81 2 address address VERB fcis-28493 81 3 this this PRON fcis-28493 81 4 , , PUNCT fcis-28493 81 5 pointtransformerconv pointtransformerconv PROPN fcis-28493 81 6 was be AUX fcis-28493 81 7 developed develop VERB fcis-28493 81 8 , , PUNCT fcis-28493 81 9 combining combine VERB fcis-28493 81 10 self self NOUN fcis-28493 81 11 - - PUNCT fcis-28493 81 12 attention attention NOUN fcis-28493 81 13 and and CCONJ fcis-28493 81 14 global global ADJ fcis-28493 81 15 - - PUNCT fcis-28493 81 16 local local ADJ fcis-28493 81 17 feature feature NOUN fcis-28493 81 18 integration integration NOUN fcis-28493 81 19 to to PART fcis-28493 81 20 significantly significantly ADV fcis-28493 81 21 improve improve VERB fcis-28493 81 22 feature feature NOUN fcis-28493 81 23 extraction extraction NOUN fcis-28493 81 24 and and CCONJ fcis-28493 81 25 model model NOUN fcis-28493 81 26 representation representation NOUN fcis-28493 81 27 . . PUNCT fcis-28493 82 1 the the DET fcis-28493 82 2 structure structure NOUN fcis-28493 82 3 of of ADP fcis-28493 82 4 pointtransformerconv pointtransformerconv NOUN fcis-28493 82 5 is be AUX fcis-28493 82 6 shown show VERB fcis-28493 82 7 in in ADP fcis-28493 82 8 fig fig NOUN fcis-28493 82 9 . . PUNCT fcis-28493 83 1 4 4 X fcis-28493 83 2 . . X fcis-28493 83 3 it it PRON fcis-28493 83 4 takes take VERB fcis-28493 83 5 the the DET fcis-28493 83 6 coordinates coordinate NOUN fcis-28493 83 7 and and CCONJ fcis-28493 83 8 positions position NOUN fcis-28493 83 9 of of ADP fcis-28493 83 10 the the DET fcis-28493 83 11 input input NOUN fcis-28493 83 12 points point NOUN fcis-28493 83 13 , , PUNCT fcis-28493 83 14 and and CCONJ fcis-28493 83 15 maps map VERB fcis-28493 83 16 them they PRON fcis-28493 83 17 into into ADP fcis-28493 83 18 a a DET fcis-28493 83 19 new new ADJ fcis-28493 83 20 feature feature NOUN fcis-28493 83 21 space space NOUN fcis-28493 83 22 using use VERB fcis-28493 83 23 two two NUM fcis-28493 83 24 linear linear ADJ fcis-28493 83 25 layers layer NOUN fcis-28493 83 26 for for ADP fcis-28493 83 27 the the DET fcis-28493 83 28 input input NOUN fcis-28493 83 29 points point NOUN fcis-28493 83 30 and and CCONJ fcis-28493 83 31 their their PRON fcis-28493 83 32 neighborhood neighborhood NOUN fcis-28493 83 33 points point NOUN fcis-28493 83 34 . . PUNCT fcis-28493 84 1 positional positional ADJ fcis-28493 84 2 encoding encoding NOUN fcis-28493 84 3 is be AUX fcis-28493 84 4 computed compute VERB fcis-28493 84 5 using use VERB fcis-28493 84 6 a a DET fcis-28493 84 7 multi multi ADJ fcis-28493 84 8 - - ADJ fcis-28493 84 9 layer layer ADJ fcis-28493 84 10 perceptron perceptron NOUN fcis-28493 84 11 ( ( PUNCT fcis-28493 84 12 mlp mlp PROPN fcis-28493 84 13 ) ) PUNCT fcis-28493 84 14 and and CCONJ fcis-28493 84 15 added add VERB fcis-28493 84 16 to to ADP fcis-28493 84 17 the the DET fcis-28493 84 18 feature feature NOUN fcis-28493 84 19 computation computation NOUN fcis-28493 84 20 , , PUNCT fcis-28493 84 21 helping help VERB fcis-28493 84 22 the the DET fcis-28493 84 23 model model NOUN fcis-28493 84 24 capture capture VERB fcis-28493 84 25 spatial spatial ADJ fcis-28493 84 26 relationships relationship NOUN fcis-28493 84 27 between between ADP fcis-28493 84 28 points point NOUN fcis-28493 84 29 . . PUNCT fcis-28493 85 1 self self NOUN fcis-28493 85 2 - - PUNCT fcis-28493 85 3 attention attention NOUN fcis-28493 85 4 scores score NOUN fcis-28493 85 5 are be AUX fcis-28493 85 6 then then ADV fcis-28493 85 7 calculated calculate VERB fcis-28493 85 8 using use VERB fcis-28493 85 9 an an DET fcis-28493 85 10 mlp mlp NOUN fcis-28493 85 11 , , PUNCT fcis-28493 85 12 with with ADP fcis-28493 85 13 the the DET fcis-28493 85 14 formula formula NOUN fcis-28493 85 15 shown show VERB fcis-28493 85 16 in in ADP fcis-28493 85 17 equation equation NOUN fcis-28493 85 18 ( ( PUNCT fcis-28493 85 19 3 3 NUM fcis-28493 85 20 ) ) PUNCT fcis-28493 85 21 . . PUNCT fcis-28493 86 1 ( ( PUNCT fcis-28493 86 2 3 3 X fcis-28493 86 3 ) ) PUNCT fcis-28493 86 4 the the DET fcis-28493 86 5 scores score NOUN fcis-28493 86 6 are be AUX fcis-28493 86 7 used use VERB fcis-28493 86 8 to to PART fcis-28493 86 9 measure measure VERB fcis-28493 86 10 the the DET fcis-28493 86 11 importance importance NOUN fcis-28493 86 12 of of ADP fcis-28493 86 13 each each DET fcis-28493 86 14 point point NOUN fcis-28493 86 15 relative relative ADJ fcis-28493 86 16 to to ADP fcis-28493 86 17 its its PRON fcis-28493 86 18 neighboring neighboring NOUN fcis-28493 86 19 points point NOUN fcis-28493 86 20 . . PUNCT fcis-28493 87 1 then then ADV fcis-28493 87 2 , , PUNCT fcis-28493 87 3 a a DET fcis-28493 87 4 linear linear ADJ fcis-28493 87 5 transformation transformation NOUN fcis-28493 87 6 is be AUX fcis-28493 87 7 applied apply VERB fcis-28493 87 8 to to ADP fcis-28493 87 9 the the DET fcis-28493 87 10 features feature NOUN fcis-28493 87 11 of of ADP fcis-28493 87 12 each each DET fcis-28493 87 13 neighboring neighboring NOUN fcis-28493 87 14 point point NOUN fcis-28493 87 15 , , PUNCT fcis-28493 87 16 along along ADP fcis-28493 87 17 with with ADP fcis-28493 87 18 the the DET fcis-28493 87 19 addition addition NOUN fcis-28493 87 20 of of ADP fcis-28493 87 21 positional positional ADJ fcis-28493 87 22 encoding encoding NOUN fcis-28493 87 23 . . PUNCT fcis-28493 88 1 the the DET fcis-28493 88 2 formula formula NOUN fcis-28493 88 3 for for ADP fcis-28493 88 4 the the DET fcis-28493 88 5 feature feature NOUN fcis-28493 88 6 linear linear PROPN fcis-28493 88 7 transformation transformation NOUN fcis-28493 88 8 is be AUX fcis-28493 88 9 shown show VERB fcis-28493 88 10 in in ADP fcis-28493 88 11 equation equation NOUN fcis-28493 88 12 ( ( PUNCT fcis-28493 88 13 4 4 NUM fcis-28493 88 14 ) ) PUNCT fcis-28493 88 15 . . PUNCT fcis-28493 89 1 ( ( PUNCT fcis-28493 89 2 4 4 X fcis-28493 89 3 ) ) PUNCT fcis-28493 89 4 finally finally ADV fcis-28493 89 5 , , PUNCT fcis-28493 89 6 the the DET fcis-28493 89 7 self self NOUN fcis-28493 89 8 - - PUNCT fcis-28493 89 9 attention attention NOUN fcis-28493 89 10 scores score NOUN fcis-28493 89 11 are be AUX fcis-28493 89 12 element element ADJ fcis-28493 89 13 - - ADJ fcis-28493 89 14 wise wise ADJ fcis-28493 89 15 multiplied multiply VERB fcis-28493 89 16 with with ADP fcis-28493 89 17 the the DET fcis-28493 89 18 transformed transform VERB fcis-28493 89 19 features feature NOUN fcis-28493 89 20 , , PUNCT fcis-28493 89 21 and and CCONJ fcis-28493 89 22 the the DET fcis-28493 89 23 results result NOUN fcis-28493 89 24 for for ADP fcis-28493 89 25 all all DET fcis-28493 89 26 points point NOUN fcis-28493 89 27 in in ADP fcis-28493 89 28 the the DET fcis-28493 89 29 neighborhood neighborhood NOUN fcis-28493 89 30 are be AUX fcis-28493 89 31 summed sum VERB fcis-28493 89 32 to to PART fcis-28493 89 33 obtain obtain VERB fcis-28493 89 34 the the DET fcis-28493 89 35 final final ADJ fcis-28493 89 36 output output NOUN fcis-28493 89 37 features feature NOUN fcis-28493 89 38 . . PUNCT fcis-28493 90 1 the the DET fcis-28493 90 2 aggregation aggregation NOUN fcis-28493 90 3 formula formula NOUN fcis-28493 90 4 is be AUX fcis-28493 90 5 shown show VERB fcis-28493 90 6 in in ADP fcis-28493 90 7 equation equation NOUN fcis-28493 90 8 ( ( PUNCT fcis-28493 90 9 5 5 NUM fcis-28493 90 10 ) ) PUNCT fcis-28493 90 11 . . PUNCT fcis-28493 91 1 ( ( PUNCT fcis-28493 91 2 5 5 X fcis-28493 91 3 ) ) PUNCT fcis-28493 91 4 fig fig NOUN fcis-28493 91 5 4 4 NUM fcis-28493 91 6 . . PUNCT fcis-28493 91 7 structure structure NOUN fcis-28493 91 8 diagram diagram NOUN fcis-28493 91 9 of of ADP fcis-28493 91 10 pointtransformerconv pointtransformerconv PROPN fcis-28493 91 11 3.4 3.4 NUM fcis-28493 91 12 . . PUNCT fcis-28493 92 1 graph graph VERB fcis-28493 92 2 neural neural ADJ fcis-28493 92 3 network network NOUN fcis-28493 92 4 module module NOUN fcis-28493 92 5 ( ( PUNCT fcis-28493 92 6 sage sage NOUN fcis-28493 92 7 ) ) PUNCT fcis-28493 92 8 graph graph NOUN fcis-28493 92 9 neural neural ADJ fcis-28493 92 10 networks network NOUN fcis-28493 92 11 ( ( PUNCT fcis-28493 92 12 gnns gnns NOUN fcis-28493 92 13 ) ) PUNCT fcis-28493 92 14 have have AUX fcis-28493 92 15 demonstrated demonstrate VERB fcis-28493 92 16 strong strong ADJ fcis-28493 92 17 expressive expressive ADJ fcis-28493 92 18 power power NOUN fcis-28493 92 19 and and CCONJ fcis-28493 92 20 adaptability adaptability NOUN fcis-28493 92 21 in in ADP fcis-28493 92 22 handling handle VERB fcis-28493 92 23 graphstructured graphstructured ADJ fcis-28493 92 24 data datum NOUN fcis-28493 92 25 . . PUNCT fcis-28493 93 1 among among ADP fcis-28493 93 2 them they PRON fcis-28493 93 3 , , PUNCT fcis-28493 93 4 sage sage NOUN fcis-28493 93 5 ( ( PUNCT fcis-28493 93 6 graphsage graphsage NOUN fcis-28493 93 7 ) ) PUNCT fcis-28493 93 8 is be AUX fcis-28493 93 9 a a DET fcis-28493 93 10 classic classic ADJ fcis-28493 93 11 graph graph NOUN fcis-28493 93 12 neural neural ADJ fcis-28493 93 13 network network NOUN fcis-28493 93 14 module module NOUN fcis-28493 93 15 that that PRON fcis-28493 93 16 effectively effectively ADV fcis-28493 93 17 improves improve VERB fcis-28493 93 18 feature feature NOUN fcis-28493 93 19 learning learning NOUN fcis-28493 93 20 and and CCONJ fcis-28493 93 21 representation representation NOUN fcis-28493 93 22 of of ADP fcis-28493 93 23 graph graph NOUN fcis-28493 93 24 data datum NOUN fcis-28493 93 25 by by ADP fcis-28493 93 26 aggregating aggregate VERB fcis-28493 93 27 information information NOUN fcis-28493 93 28 from from ADP fcis-28493 93 29 neighboring neighboring NOUN fcis-28493 93 30 nodes node NOUN fcis-28493 93 31 . . PUNCT fcis-28493 94 1 sage sage NOUN fcis-28493 94 2 updates update VERB fcis-28493 94 3 node node VERB fcis-28493 94 4 feature feature NOUN fcis-28493 94 5 representations representation NOUN fcis-28493 94 6 through through ADP fcis-28493 94 7 multilayer multilayer ADJ fcis-28493 94 8 aggregation aggregation NOUN fcis-28493 94 9 of of ADP fcis-28493 94 10 neighborhood neighborhood NOUN fcis-28493 94 11 node node ADJ fcis-28493 94 12 information information NOUN fcis-28493 94 13 . . PUNCT fcis-28493 95 1 in in ADP fcis-28493 95 2 this this DET fcis-28493 95 3 paper paper NOUN fcis-28493 95 4 , , PUNCT fcis-28493 95 5 the the DET fcis-28493 95 6 sage sage NOUN fcis-28493 95 7 module module NOUN fcis-28493 95 8 is be AUX fcis-28493 95 9 used use VERB fcis-28493 95 10 for for ADP fcis-28493 95 11 feature feature NOUN fcis-28493 95 12 extraction extraction NOUN fcis-28493 95 13 from from ADP fcis-28493 95 14 point point NOUN fcis-28493 95 15 cloud cloud NOUN fcis-28493 95 16 data datum NOUN fcis-28493 95 17 , , PUNCT fcis-28493 95 18 aggregating aggregate VERB fcis-28493 95 19 local local ADJ fcis-28493 95 20 features feature NOUN fcis-28493 95 21 such such ADJ fcis-28493 95 22 as as ADP fcis-28493 95 23 position position NOUN fcis-28493 95 24 and and CCONJ fcis-28493 95 25 normal normal ADJ fcis-28493 95 26 vectors vector NOUN fcis-28493 95 27 , , PUNCT fcis-28493 95 28 helping help VERB fcis-28493 95 29 the the DET fcis-28493 95 30 model model NOUN fcis-28493 95 31 understand understand VERB fcis-28493 95 32 the the DET fcis-28493 95 33 local local ADJ fcis-28493 95 34 neighborhood neighborhood NOUN fcis-28493 95 35 structure structure NOUN fcis-28493 95 36 of of ADP fcis-28493 95 37 each each DET fcis-28493 95 38 point point NOUN fcis-28493 95 39 . . PUNCT fcis-28493 96 1 this this PRON fcis-28493 96 2 provides provide VERB fcis-28493 96 3 a a DET fcis-28493 96 4 foundation foundation NOUN fcis-28493 96 5 for for ADP fcis-28493 96 6 subsequent subsequent ADJ fcis-28493 96 7 global global ADJ fcis-28493 96 8 feature feature NOUN fcis-28493 96 9 integration integration NOUN fcis-28493 96 10 , , PUNCT fcis-28493 96 11 enabling enable VERB fcis-28493 96 12 the the DET fcis-28493 96 13 model model NOUN fcis-28493 96 14 to to PART fcis-28493 96 15 more more ADV fcis-28493 96 16 accurately accurately ADV fcis-28493 96 17 assess assess VERB fcis-28493 96 18 the the DET fcis-28493 96 19 role role NOUN fcis-28493 96 20 of of ADP fcis-28493 96 21 each each DET fcis-28493 96 22 point point NOUN fcis-28493 96 23 in in ADP fcis-28493 96 24 the the DET fcis-28493 96 25 grasping grasp VERB fcis-28493 96 26 task task NOUN fcis-28493 96 27 . . PUNCT fcis-28493 97 1 compared compare VERB fcis-28493 97 2 to to ADP fcis-28493 97 3 traditional traditional ADJ fcis-28493 97 4 methods method NOUN fcis-28493 97 5 like like ADP fcis-28493 97 6 pointnetconv pointnetconv NOUN fcis-28493 97 7 , , PUNCT fcis-28493 97 8 sage sage NOUN fcis-28493 97 9 enhances enhance VERB fcis-28493 97 10 the the DET fcis-28493 97 11 model model NOUN fcis-28493 97 12 ’s ’s PART fcis-28493 97 13 expressive expressive ADJ fcis-28493 97 14 power power NOUN fcis-28493 97 15 and and CCONJ fcis-28493 97 16 task task NOUN fcis-28493 97 17 performance performance NOUN fcis-28493 97 18 through through ADP fcis-28493 97 19 multi multi ADJ fcis-28493 97 20 - - ADJ fcis-28493 97 21 layer layer ADJ fcis-28493 97 22 information information NOUN fcis-28493 97 23 propagation propagation NOUN fcis-28493 97 24 and and CCONJ fcis-28493 97 25 feature feature NOUN fcis-28493 97 26 aggregation aggregation NOUN fcis-28493 97 27 . . PUNCT fcis-28493 98 1 the the DET fcis-28493 98 2 structure structure NOUN fcis-28493 98 3 of of ADP fcis-28493 98 4 sage sage NOUN fcis-28493 98 5 is be AUX fcis-28493 98 6 shown show VERB fcis-28493 98 7 in in ADP fcis-28493 98 8 fig fig NOUN fcis-28493 98 9 . . PUNCT fcis-28493 99 1 5 5 NUM fcis-28493 99 2 , , PUNCT fcis-28493 99 3 and and CCONJ fcis-28493 99 4 its its PRON fcis-28493 99 5 architecture architecture NOUN fcis-28493 99 6 involves involve VERB fcis-28493 99 7 sampling sample VERB fcis-28493 99 8 neighborhood neighborhood NOUN fcis-28493 99 9 nodes node NOUN fcis-28493 99 10 after after ADP fcis-28493 99 11 the the DET fcis-28493 99 12 input input NOUN fcis-28493 99 13 node node NOUN fcis-28493 99 14 features feature VERB fcis-28493 99 15 . . PUNCT fcis-28493 100 1 the the DET fcis-28493 100 2 sampling sample VERB fcis-28493 100 3 formula formula NOUN fcis-28493 100 4 is be AUX fcis-28493 100 5 shown show VERB fcis-28493 100 6 in in ADP fcis-28493 100 7 equation equation NOUN fcis-28493 100 8 ( ( PUNCT fcis-28493 100 9 6 6 NUM fcis-28493 100 10 ) ) PUNCT fcis-28493 100 11 . . PUNCT fcis-28493 101 1 ( ( PUNCT fcis-28493 101 2 6 6 NUM fcis-28493 101 3 ) ) PUNCT fcis-28493 101 4 here here ADV fcis-28493 101 5 , , PUNCT fcis-28493 101 6 represents represent VERB fcis-28493 101 7 the the DET fcis-28493 101 8 set set NOUN fcis-28493 101 9 of of ADP fcis-28493 101 10 neighboring neighboring NOUN fcis-28493 101 11 nodes node NOUN fcis-28493 101 12 of of ADP fcis-28493 101 13 node node NOUN fcis-28493 101 14 . . PUNCT fcis-28493 102 1 in in ADP fcis-28493 102 2 this this DET fcis-28493 102 3 paper paper NOUN fcis-28493 102 4 , , PUNCT fcis-28493 102 5 the the DET fcis-28493 102 6 aggregation aggregation NOUN fcis-28493 102 7 is be AUX fcis-28493 102 8 performed perform VERB fcis-28493 102 9 by by ADP fcis-28493 102 10 taking take VERB fcis-28493 102 11 the the DET fcis-28493 102 12 mean mean NOUN fcis-28493 102 13 of of ADP fcis-28493 102 14 the the DET fcis-28493 102 15 features feature NOUN fcis-28493 102 16 . . PUNCT fcis-28493 103 1 the the DET fcis-28493 103 2 aggregation aggregation NOUN fcis-28493 103 3 formula formula NOUN fcis-28493 103 4 is be AUX fcis-28493 103 5 shown show VERB fcis-28493 103 6 in in ADP fcis-28493 103 7 equation equation NOUN fcis-28493 103 8 ( ( PUNCT fcis-28493 103 9 7 7 NUM fcis-28493 103 10 ) ) PUNCT fcis-28493 103 11 . . PUNCT fcis-28493 104 1 ( ( PUNCT fcis-28493 104 2 7 7 X fcis-28493 104 3 ) ) PUNCT fcis-28493 104 4 after after ADP fcis-28493 104 5 aggregation aggregation NOUN fcis-28493 104 6 , , PUNCT fcis-28493 104 7 the the DET fcis-28493 104 8 node node NOUN fcis-28493 104 9 features feature NOUN fcis-28493 104 10 are be AUX fcis-28493 104 11 updated update VERB fcis-28493 104 12 through through ADP fcis-28493 104 13 an an DET fcis-28493 104 14 mlp mlp NOUN fcis-28493 104 15 . . PUNCT fcis-28493 105 1 the the DET fcis-28493 105 2 formula formula NOUN fcis-28493 105 3 for for ADP fcis-28493 105 4 updating update VERB fcis-28493 105 5 the the DET fcis-28493 105 6 node node NOUN fcis-28493 105 7 features feature NOUN fcis-28493 105 8 is be AUX fcis-28493 105 9 shown show VERB fcis-28493 105 10 in in ADP fcis-28493 105 11 equation equation NOUN fcis-28493 105 12 ( ( PUNCT fcis-28493 105 13 8) 8) NUM fcis-28493 105 14 . . PUNCT fcis-28493 106 1 ( ( PUNCT fcis-28493 106 2 8) 8) NUM fcis-28493 106 3 4 4 NUM fcis-28493 106 4 . . PUNCT fcis-28493 106 5 dataset dataset NOUN fcis-28493 106 6 and and CCONJ fcis-28493 106 7 data datum NOUN fcis-28493 106 8 preprocessing preprocesse VERB fcis-28493 106 9 4.1 4.1 NUM fcis-28493 106 10 . . PUNCT fcis-28493 107 1 dataset dataset VERB fcis-28493 107 2 in in ADP fcis-28493 107 3 this this DET fcis-28493 107 4 paper paper NOUN fcis-28493 107 5 , , PUNCT fcis-28493 107 6 the the DET fcis-28493 107 7 ycb ycb NOUN fcis-28493 107 8 , , PUNCT fcis-28493 107 9 bigbird bigbird PROPN fcis-28493 107 10 , , PUNCT fcis-28493 107 11 and and CCONJ fcis-28493 107 12 kit kit NOUN fcis-28493 107 13 datasets dataset NOUN fcis-28493 107 14 were be AUX fcis-28493 107 15 68 68 NUM fcis-28493 107 16 chosen choose VERB fcis-28493 107 17 for for ADP fcis-28493 107 18 model model NOUN fcis-28493 107 19 training training NOUN fcis-28493 107 20 experiments[11 experiments[11 NOUN fcis-28493 107 21 ] ] PUNCT fcis-28493 107 22 . . PUNCT fcis-28493 108 1 the the DET fcis-28493 108 2 training training NOUN fcis-28493 108 3 set set NOUN fcis-28493 108 4 contains contain VERB fcis-28493 108 5 200 200 NUM fcis-28493 108 6 objects object NOUN fcis-28493 108 7 , , PUNCT fcis-28493 108 8 while while SCONJ fcis-28493 108 9 the the DET fcis-28493 108 10 test test NOUN fcis-28493 108 11 set set NOUN fcis-28493 108 12 contains contain VERB fcis-28493 108 13 40 40 NUM fcis-28493 108 14 objects object NOUN fcis-28493 108 15 . . PUNCT fcis-28493 109 1 to to PART fcis-28493 109 2 effectively effectively ADV fcis-28493 109 3 utilize utilize VERB fcis-28493 109 4 point point NOUN fcis-28493 109 5 cloud cloud NOUN fcis-28493 109 6 data datum NOUN fcis-28493 109 7 for for ADP fcis-28493 109 8 model model NOUN fcis-28493 109 9 training training NOUN fcis-28493 109 10 , , PUNCT fcis-28493 109 11 key key ADJ fcis-28493 109 12 features feature NOUN fcis-28493 109 13 such such ADJ fcis-28493 109 14 as as ADP fcis-28493 109 15 point point NOUN fcis-28493 109 16 cloud cloud NOUN fcis-28493 109 17 coordinates coordinate NOUN fcis-28493 109 18 and and CCONJ fcis-28493 109 19 normal normal ADJ fcis-28493 109 20 information information NOUN fcis-28493 109 21 were be AUX fcis-28493 109 22 first first ADV fcis-28493 109 23 extracted extract VERB fcis-28493 109 24 from from ADP fcis-28493 109 25 the the DET fcis-28493 109 26 .npz .npz NOUN fcis-28493 109 27 files file NOUN fcis-28493 109 28 . . PUNCT fcis-28493 110 1 the the DET fcis-28493 110 2 data datum NOUN fcis-28493 110 3 was be AUX fcis-28493 110 4 then then ADV fcis-28493 110 5 processed process VERB fcis-28493 110 6 using use VERB fcis-28493 110 7 a a DET fcis-28493 110 8 custom custom NOUN fcis-28493 110 9 dataset dataset NOUN fcis-28493 110 10 class class NOUN fcis-28493 110 11 , , PUNCT fcis-28493 110 12 filtering filter VERB fcis-28493 110 13 out out ADP fcis-28493 110 14 point point NOUN fcis-28493 110 15 clouds cloud NOUN fcis-28493 110 16 with with ADP fcis-28493 110 17 fewer few ADJ fcis-28493 110 18 than than ADP fcis-28493 110 19 10 10 NUM fcis-28493 110 20 labels label NOUN fcis-28493 110 21 or or CCONJ fcis-28493 110 22 those those PRON fcis-28493 110 23 consisting consist VERB fcis-28493 110 24 entirely entirely ADV fcis-28493 110 25 of of ADP fcis-28493 110 26 positive positive ADJ fcis-28493 110 27 samples sample NOUN fcis-28493 110 28 , , PUNCT fcis-28493 110 29 ensuring ensure VERB fcis-28493 110 30 data datum NOUN fcis-28493 110 31 diversity diversity NOUN fcis-28493 110 32 . . PUNCT fcis-28493 111 1 during during ADP fcis-28493 111 2 the the DET fcis-28493 111 3 preprocessing preprocessing NOUN fcis-28493 111 4 stage stage NOUN fcis-28493 111 5 , , PUNCT fcis-28493 111 6 the the DET fcis-28493 111 7 data datum NOUN fcis-28493 111 8 underwent underwent NOUN fcis-28493 111 9 centering center VERB fcis-28493 111 10 and and CCONJ fcis-28493 111 11 random random ADJ fcis-28493 111 12 rotation rotation NOUN fcis-28493 111 13 augmentation augmentation NOUN fcis-28493 111 14 . . PUNCT fcis-28493 112 1 finally finally ADV fcis-28493 112 2 , , PUNCT fcis-28493 112 3 the the DET fcis-28493 112 4 data datum NOUN fcis-28493 112 5 was be AUX fcis-28493 112 6 subsampled subsample VERB fcis-28493 112 7 , , PUNCT fcis-28493 112 8 divided divide VERB fcis-28493 112 9 into into ADP fcis-28493 112 10 training training NOUN fcis-28493 112 11 and and CCONJ fcis-28493 112 12 test test NOUN fcis-28493 112 13 sets set NOUN fcis-28493 112 14 , , PUNCT fcis-28493 112 15 and and CCONJ fcis-28493 112 16 saved save VERB fcis-28493 112 17 in in ADP fcis-28493 112 18 pytorch pytorch NOUN fcis-28493 112 19 format format NOUN fcis-28493 112 20 for for ADP fcis-28493 112 21 subsequent subsequent ADJ fcis-28493 112 22 model model NOUN fcis-28493 112 23 training training NOUN fcis-28493 112 24 and and CCONJ fcis-28493 112 25 evaluation evaluation NOUN fcis-28493 112 26 . . PUNCT fcis-28493 113 1 fig fig NOUN fcis-28493 113 2 5 5 NUM fcis-28493 113 3 . . PUNCT fcis-28493 114 1 graphsage graphsage NOUN fcis-28493 114 2 principle principle ADJ fcis-28493 114 3 diagram diagram NOUN fcis-28493 114 4 4.2 4.2 NUM fcis-28493 114 5 . . PUNCT fcis-28493 115 1 data datum NOUN fcis-28493 115 2 preprocessing preprocessing NOUN fcis-28493 115 3 during during ADP fcis-28493 115 4 the the DET fcis-28493 115 5 data datum NOUN fcis-28493 115 6 preprocessing preprocessing NOUN fcis-28493 115 7 stage stage NOUN fcis-28493 115 8 , , PUNCT fcis-28493 115 9 this this DET fcis-28493 115 10 paper paper NOUN fcis-28493 115 11 performed perform VERB fcis-28493 115 12 centering center VERB fcis-28493 115 13 and and CCONJ fcis-28493 115 14 random random ADJ fcis-28493 115 15 rotation rotation NOUN fcis-28493 115 16 augmentation augmentation NOUN fcis-28493 115 17 on on ADP fcis-28493 115 18 the the DET fcis-28493 115 19 point point NOUN fcis-28493 115 20 cloud cloud NOUN fcis-28493 115 21 data datum NOUN fcis-28493 115 22 to to PART fcis-28493 115 23 improve improve VERB fcis-28493 115 24 data datum NOUN fcis-28493 115 25 quality quality NOUN fcis-28493 115 26 and and CCONJ fcis-28493 115 27 enhance enhance VERB fcis-28493 115 28 the the DET fcis-28493 115 29 model model NOUN fcis-28493 115 30 's 's PART fcis-28493 115 31 generalization generalization NOUN fcis-28493 115 32 ability ability NOUN fcis-28493 115 33 . . PUNCT fcis-28493 116 1 centering center VERB fcis-28493 116 2 : : PUNCT fcis-28493 116 3 to to PART fcis-28493 116 4 ensure ensure VERB fcis-28493 116 5 that that SCONJ fcis-28493 116 6 the the DET fcis-28493 116 7 center center NOUN fcis-28493 116 8 of of ADP fcis-28493 116 9 the the DET fcis-28493 116 10 point point NOUN fcis-28493 116 11 cloud cloud NOUN fcis-28493 116 12 is be AUX fcis-28493 116 13 at at ADP fcis-28493 116 14 the the DET fcis-28493 116 15 coordinate coordinate ADJ fcis-28493 116 16 origin origin NOUN fcis-28493 116 17 and and CCONJ fcis-28493 116 18 reduce reduce VERB fcis-28493 116 19 training training NOUN fcis-28493 116 20 bias bias NOUN fcis-28493 116 21 caused cause VERB fcis-28493 116 22 by by ADP fcis-28493 116 23 variations variation NOUN fcis-28493 116 24 in in ADP fcis-28493 116 25 the the DET fcis-28493 116 26 point point NOUN fcis-28493 116 27 cloud cloud NOUN fcis-28493 116 28 ’s ’s PART fcis-28493 116 29 position position NOUN fcis-28493 116 30 , , PUNCT fcis-28493 116 31 the the DET fcis-28493 116 32 centroid centroid NOUN fcis-28493 116 33 of of ADP fcis-28493 116 34 the the DET fcis-28493 116 35 point point NOUN fcis-28493 116 36 cloud cloud NOUN fcis-28493 116 37 data datum NOUN fcis-28493 116 38 was be AUX fcis-28493 116 39 first first ADV fcis-28493 116 40 calculated calculate VERB fcis-28493 116 41 , , PUNCT fcis-28493 116 42 which which PRON fcis-28493 116 43 is be AUX fcis-28493 116 44 the the DET fcis-28493 116 45 average average NOUN fcis-28493 116 46 of of ADP fcis-28493 116 47 the the DET fcis-28493 116 48 coordinates coordinate NOUN fcis-28493 116 49 of of ADP fcis-28493 116 50 all all DET fcis-28493 116 51 points point NOUN fcis-28493 116 52 . . PUNCT fcis-28493 117 1 the the DET fcis-28493 117 2 centroid centroid NOUN fcis-28493 117 3 was be AUX fcis-28493 117 4 then then ADV fcis-28493 117 5 subtracted subtract VERB fcis-28493 117 6 from from ADP fcis-28493 117 7 the the DET fcis-28493 117 8 position position NOUN fcis-28493 117 9 of of ADP fcis-28493 117 10 each each DET fcis-28493 117 11 point point NOUN fcis-28493 117 12 , , PUNCT fcis-28493 117 13 achieving achieve VERB fcis-28493 117 14 centering center VERB fcis-28493 117 15 of of ADP fcis-28493 117 16 the the DET fcis-28493 117 17 point point NOUN fcis-28493 117 18 cloud cloud NOUN fcis-28493 117 19 . . PUNCT fcis-28493 118 1 the the DET fcis-28493 118 2 formula formula NOUN fcis-28493 118 3 for for ADP fcis-28493 118 4 calculating calculate VERB fcis-28493 118 5 the the DET fcis-28493 118 6 centroid centroid NOUN fcis-28493 118 7 is be AUX fcis-28493 118 8 shown show VERB fcis-28493 118 9 in in ADP fcis-28493 118 10 equation equation NOUN fcis-28493 118 11 ( ( PUNCT fcis-28493 118 12 9 9 NUM fcis-28493 118 13 ) ) PUNCT fcis-28493 118 14 . . PUNCT fcis-28493 119 1 ( ( PUNCT fcis-28493 119 2 9 9 X fcis-28493 119 3 ) ) PUNCT fcis-28493 119 4 here here ADV fcis-28493 119 5 , , PUNCT fcis-28493 119 6 is be AUX fcis-28493 119 7 the the DET fcis-28493 119 8 number number NOUN fcis-28493 119 9 of of ADP fcis-28493 119 10 points point NOUN fcis-28493 119 11 , , PUNCT fcis-28493 119 12 and and CCONJ fcis-28493 119 13 is be AUX fcis-28493 119 14 the the DET fcis-28493 119 15 position position NOUN fcis-28493 119 16 of of ADP fcis-28493 119 17 the the DET fcis-28493 119 18 i i PROPN fcis-28493 119 19 - - PUNCT fcis-28493 119 20 th th VERB fcis-28493 119 21 point point NOUN fcis-28493 119 22 . . PUNCT fcis-28493 120 1 random random ADJ fcis-28493 120 2 rotation rotation NOUN fcis-28493 120 3 augmentation augmentation NOUN fcis-28493 120 4 : : PUNCT fcis-28493 120 5 random random ADJ fcis-28493 120 6 rotation rotation NOUN fcis-28493 120 7 augmentation augmentation NOUN fcis-28493 120 8 is be AUX fcis-28493 120 9 applied apply VERB fcis-28493 120 10 to to ADP fcis-28493 120 11 the the DET fcis-28493 120 12 point point NOUN fcis-28493 120 13 cloud cloud NOUN fcis-28493 120 14 data datum NOUN fcis-28493 120 15 to to PART fcis-28493 120 16 improve improve VERB fcis-28493 120 17 the the DET fcis-28493 120 18 model model NOUN fcis-28493 120 19 's 's PART fcis-28493 120 20 robustness robustness NOUN fcis-28493 120 21 . . PUNCT fcis-28493 121 1 first first ADV fcis-28493 121 2 , , PUNCT fcis-28493 121 3 a a DET fcis-28493 121 4 rotation rotation NOUN fcis-28493 121 5 matrix matrix NOUN fcis-28493 121 6 is be AUX fcis-28493 121 7 generated generate VERB fcis-28493 121 8 by by ADP fcis-28493 121 9 randomly randomly ADV fcis-28493 121 10 selecting select VERB fcis-28493 121 11 rotation rotation NOUN fcis-28493 121 12 angles angle NOUN fcis-28493 121 13 in in ADP fcis-28493 121 14 3d 3d NUM fcis-28493 121 15 space space NOUN fcis-28493 121 16 . . PUNCT fcis-28493 122 1 the the DET fcis-28493 122 2 rotation rotation NOUN fcis-28493 122 3 matrix matrix NOUN fcis-28493 122 4 is be AUX fcis-28493 122 5 composed compose VERB fcis-28493 122 6 of of ADP fcis-28493 122 7 rotation rotation NOUN fcis-28493 122 8 matrices matrix NOUN fcis-28493 122 9 around around ADP fcis-28493 122 10 the the DET fcis-28493 122 11 x x PROPN fcis-28493 122 12 , , PUNCT fcis-28493 122 13 y y PROPN fcis-28493 122 14 , , PUNCT fcis-28493 122 15 and and CCONJ fcis-28493 122 16 z z NOUN fcis-28493 122 17 axes axis NOUN fcis-28493 122 18 , , PUNCT fcis-28493 122 19 as as SCONJ fcis-28493 122 20 shown show VERB fcis-28493 122 21 in in ADP fcis-28493 122 22 equations equation NOUN fcis-28493 122 23 ( ( PUNCT fcis-28493 122 24 10)-(12 10)-(12 NUM fcis-28493 122 25 ) ) PUNCT fcis-28493 122 26 . . PUNCT fcis-28493 123 1 ( ( PUNCT fcis-28493 123 2 10 10 NUM fcis-28493 123 3 ) ) PUNCT fcis-28493 123 4 ( ( PUNCT fcis-28493 123 5 11 11 NUM fcis-28493 123 6 ) ) PUNCT fcis-28493 123 7 ( ( PUNCT fcis-28493 123 8 12 12 NUM fcis-28493 123 9 ) ) PUNCT fcis-28493 123 10 the the DET fcis-28493 123 11 rotation rotation NOUN fcis-28493 123 12 matrix matrix NOUN fcis-28493 123 13 is be AUX fcis-28493 123 14 the the DET fcis-28493 123 15 product product NOUN fcis-28493 123 16 of of ADP fcis-28493 123 17 the the DET fcis-28493 123 18 rotation rotation NOUN fcis-28493 123 19 matrices matrix NOUN fcis-28493 123 20 mentioned mention VERB fcis-28493 123 21 above above ADV fcis-28493 123 22 . . PUNCT fcis-28493 124 1 using use VERB fcis-28493 124 2 this this DET fcis-28493 124 3 rotation rotation NOUN fcis-28493 124 4 matrix matrix NOUN fcis-28493 124 5 , , PUNCT fcis-28493 124 6 the the DET fcis-28493 124 7 point point NOUN fcis-28493 124 8 cloud cloud NOUN fcis-28493 124 9 data datum NOUN fcis-28493 124 10 can can AUX fcis-28493 124 11 be be AUX fcis-28493 124 12 rotated rotate VERB fcis-28493 124 13 . . PUNCT fcis-28493 125 1 this this DET fcis-28493 125 2 method method NOUN fcis-28493 125 3 allows allow VERB fcis-28493 125 4 the the DET fcis-28493 125 5 point point NOUN fcis-28493 125 6 cloud cloud NOUN fcis-28493 125 7 data datum NOUN fcis-28493 125 8 to to PART fcis-28493 125 9 be be AUX fcis-28493 125 10 rotated rotate VERB fcis-28493 125 11 to to ADP fcis-28493 125 12 different different ADJ fcis-28493 125 13 angles angle NOUN fcis-28493 125 14 , , PUNCT fcis-28493 125 15 thereby thereby ADV fcis-28493 125 16 increasing increase VERB fcis-28493 125 17 the the DET fcis-28493 125 18 diversity diversity NOUN fcis-28493 125 19 of of ADP fcis-28493 125 20 the the DET fcis-28493 125 21 training training NOUN fcis-28493 125 22 data datum NOUN fcis-28493 125 23 and and CCONJ fcis-28493 125 24 enhancing enhance VERB fcis-28493 125 25 the the DET fcis-28493 125 26 model model NOUN fcis-28493 125 27 's 's PART fcis-28493 125 28 robustness robustness NOUN fcis-28493 125 29 . . PUNCT fcis-28493 126 1 5 5 X fcis-28493 126 2 . . PUNCT fcis-28493 126 3 model model NOUN fcis-28493 126 4 training training NOUN fcis-28493 126 5 and and CCONJ fcis-28493 126 6 comparison comparison NOUN fcis-28493 126 7 experiments experiment NOUN fcis-28493 126 8 5.1 5.1 NUM fcis-28493 126 9 . . PUNCT fcis-28493 127 1 model model NOUN fcis-28493 127 2 training train VERB fcis-28493 127 3 the the DET fcis-28493 127 4 experimental experimental ADJ fcis-28493 127 5 simulation simulation NOUN fcis-28493 127 6 environment environment NOUN fcis-28493 127 7 in in ADP fcis-28493 127 8 this this DET fcis-28493 127 9 paper paper NOUN fcis-28493 127 10 was be AUX fcis-28493 127 11 run run VERB fcis-28493 127 12 on on ADP fcis-28493 127 13 an an DET fcis-28493 127 14 nvidia nvidia PROPN fcis-28493 127 15 l20 l20 PROPN fcis-28493 127 16 gpu gpu PROPN fcis-28493 127 17 server server NOUN fcis-28493 127 18 , , PUNCT fcis-28493 127 19 using use VERB fcis-28493 127 20 the the DET fcis-28493 127 21 pytorch pytorch NOUN fcis-28493 127 22 neural neural ADJ fcis-28493 127 23 network network NOUN fcis-28493 127 24 framework framework NOUN fcis-28493 127 25 . . PUNCT fcis-28493 128 1 training training NOUN fcis-28493 128 2 data datum NOUN fcis-28493 128 3 was be AUX fcis-28493 128 4 created create VERB fcis-28493 128 5 by by ADP fcis-28493 128 6 simulating simulate VERB fcis-28493 128 7 vertical vertical ADJ fcis-28493 128 8 and and CCONJ fcis-28493 128 9 random random ADJ fcis-28493 128 10 placement placement NOUN fcis-28493 128 11 environments environment NOUN fcis-28493 128 12 , , PUNCT fcis-28493 128 13 where where SCONJ fcis-28493 128 14 each each DET fcis-28493 128 15 scene scene NOUN fcis-28493 128 16 contains contain VERB fcis-28493 128 17 a a DET fcis-28493 128 18 random random ADJ fcis-28493 128 19 number number NOUN fcis-28493 128 20 of of ADP fcis-28493 128 21 objects object NOUN fcis-28493 128 22 . . PUNCT fcis-28493 129 1 depth depth NOUN fcis-28493 129 2 images image NOUN fcis-28493 129 3 captured capture VERB fcis-28493 129 4 from from ADP fcis-28493 129 5 different different ADJ fcis-28493 129 6 camera camera NOUN fcis-28493 129 7 viewpoints viewpoint NOUN fcis-28493 129 8 were be AUX fcis-28493 129 9 corrupted corrupt VERB fcis-28493 129 10 with with ADP fcis-28493 129 11 gaussian gaussian ADJ fcis-28493 129 12 noise noise NOUN fcis-28493 129 13 , , PUNCT fcis-28493 129 14 voxelized voxelize VERB fcis-28493 129 15 , , PUNCT fcis-28493 129 16 and and CCONJ fcis-28493 129 17 up up ADP fcis-28493 129 18 to to PART fcis-28493 129 19 2000 2000 NUM fcis-28493 129 20 edge edge NOUN fcis-28493 129 21 grasping grasp VERB fcis-28493 129 22 candidates candidate NOUN fcis-28493 129 23 were be AUX fcis-28493 129 24 generated generate VERB fcis-28493 129 25 for for ADP fcis-28493 129 26 each each DET fcis-28493 129 27 scene scene NOUN fcis-28493 129 28 . . PUNCT fcis-28493 130 1 the the DET fcis-28493 130 2 grasping grasp VERB fcis-28493 130 3 candidates candidate NOUN fcis-28493 130 4 were be AUX fcis-28493 130 5 labeled label VERB fcis-28493 130 6 through through ADP fcis-28493 130 7 simulated simulate VERB fcis-28493 130 8 grasping grasp VERB fcis-28493 130 9 attempts attempt NOUN fcis-28493 130 10 . . PUNCT fcis-28493 131 1 when when SCONJ fcis-28493 131 2 generating generate VERB fcis-28493 131 3 2000 2000 NUM fcis-28493 131 4 candidates candidate NOUN fcis-28493 131 5 , , PUNCT fcis-28493 131 6 32 32 NUM fcis-28493 131 7 neighboring neighboring NOUN fcis-28493 131 8 points point NOUN fcis-28493 131 9 were be AUX fcis-28493 131 10 randomly randomly ADV fcis-28493 131 11 sampled sample VERB fcis-28493 131 12 from from ADP fcis-28493 131 13 the the DET fcis-28493 131 14 voxelized voxelize VERB fcis-28493 131 15 point point NOUN fcis-28493 131 16 cloud cloud NOUN fcis-28493 131 17 . . PUNCT fcis-28493 132 1 model model NOUN fcis-28493 132 2 training training NOUN fcis-28493 132 3 used use VERB fcis-28493 132 4 the the DET fcis-28493 132 5 adam adam PROPN fcis-28493 132 6 optimizer optimizer NOUN fcis-28493 132 7 with with ADP fcis-28493 132 8 an an DET fcis-28493 132 9 initial initial ADJ fcis-28493 132 10 learning learning NOUN fcis-28493 132 11 rate rate NOUN fcis-28493 132 12 of of ADP fcis-28493 132 13 η\etaη η\etaη NOUN fcis-28493 132 14 , , PUNCT fcis-28493 132 15 and and CCONJ fcis-28493 132 16 the the DET fcis-28493 132 17 learning learning NOUN fcis-28493 132 18 rate rate NOUN fcis-28493 132 19 was be AUX fcis-28493 132 20 halved halve VERB fcis-28493 132 21 after after ADP fcis-28493 132 22 6 6 NUM fcis-28493 132 23 epochs epoch NOUN fcis-28493 132 24 of of ADP fcis-28493 132 25 stagnation stagnation NOUN fcis-28493 132 26 in in ADP fcis-28493 132 27 the the DET fcis-28493 132 28 test test NOUN fcis-28493 132 29 loss loss NOUN fcis-28493 132 30 . . PUNCT fcis-28493 133 1 5.2 5.2 NUM fcis-28493 133 2 . . PUNCT fcis-28493 134 1 comparison comparison NOUN fcis-28493 134 2 experiments experiment NOUN fcis-28493 134 3 fig fig NOUN fcis-28493 134 4 6 6 NUM fcis-28493 134 5 . . PUNCT fcis-28493 135 1 algorithm algorithm PROPN fcis-28493 135 2 comparison comparison NOUN fcis-28493 135 3 diagram diagram PROPN fcis-28493 135 4 69 69 NUM fcis-28493 135 5 to to PART fcis-28493 135 6 objectively objectively ADV fcis-28493 135 7 evaluate evaluate VERB fcis-28493 135 8 the the DET fcis-28493 135 9 performance performance NOUN fcis-28493 135 10 of of ADP fcis-28493 135 11 the the DET fcis-28493 135 12 improved improved ADJ fcis-28493 135 13 network network NOUN fcis-28493 135 14 , , PUNCT fcis-28493 135 15 this this DET fcis-28493 135 16 paper paper NOUN fcis-28493 135 17 uses use VERB fcis-28493 135 18 test test NOUN fcis-28493 135 19 accuracy accuracy NOUN fcis-28493 135 20 ( ( PUNCT fcis-28493 135 21 test_acc test_acc NOUN fcis-28493 135 22 ) ) PUNCT fcis-28493 135 23 and and CCONJ fcis-28493 135 24 loss loss NOUN fcis-28493 135 25 value value NOUN fcis-28493 135 26 as as ADP fcis-28493 135 27 the the DET fcis-28493 135 28 evaluation evaluation NOUN fcis-28493 135 29 metrics metric NOUN fcis-28493 135 30 for for ADP fcis-28493 135 31 the the DET fcis-28493 135 32 model model NOUN fcis-28493 135 33 . . PUNCT fcis-28493 136 1 to to PART fcis-28493 136 2 verify verify VERB fcis-28493 136 3 the the DET fcis-28493 136 4 superiority superiority NOUN fcis-28493 136 5 of of ADP fcis-28493 136 6 the the DET fcis-28493 136 7 proposed propose VERB fcis-28493 136 8 tes tes PROPN fcis-28493 136 9 - - PUNCT fcis-28493 136 10 net net ADJ fcis-28493 136 11 model model NOUN fcis-28493 136 12 , , PUNCT fcis-28493 136 13 comparison comparison NOUN fcis-28493 136 14 experiments experiment NOUN fcis-28493 136 15 were be AUX fcis-28493 136 16 conducted conduct VERB fcis-28493 136 17 with with ADP fcis-28493 136 18 the the DET fcis-28493 136 19 current current ADJ fcis-28493 136 20 mainstream mainstream NOUN fcis-28493 136 21 algorithms algorithm NOUN fcis-28493 136 22 : : PUNCT fcis-28493 136 23 pointnetgpd pointnetgpd PROPN fcis-28493 136 24 , , PUNCT fcis-28493 136 25 gpd gpd PROPN fcis-28493 136 26 , , PUNCT fcis-28493 136 27 the the DET fcis-28493 136 28 original original ADJ fcis-28493 136 29 edgegrasp edgegrasp NOUN fcis-28493 136 30 - - PUNCT fcis-28493 136 31 net net NOUN fcis-28493 136 32 algorithm algorithm NOUN fcis-28493 136 33 , , PUNCT fcis-28493 136 34 and and CCONJ fcis-28493 136 35 a a DET fcis-28493 136 36 version version NOUN fcis-28493 136 37 of of ADP fcis-28493 136 38 tes tes NOUN fcis-28493 136 39 - - NOUN fcis-28493 136 40 net net NOUN fcis-28493 136 41 with with ADP fcis-28493 136 42 the the DET fcis-28493 136 43 graph graph NOUN fcis-28493 136 44 neural neural ADJ fcis-28493 136 45 network network NOUN fcis-28493 136 46 module module NOUN fcis-28493 136 47 replaced replace VERB fcis-28493 136 48 by by ADP fcis-28493 136 49 the the DET fcis-28493 136 50 gat2v gat2v PROPN fcis-28493 136 51 module module NOUN fcis-28493 136 52 . . PUNCT fcis-28493 137 1 this this DET fcis-28493 137 2 experiment experiment NOUN fcis-28493 137 3 primarily primarily ADV fcis-28493 137 4 compares compare VERB fcis-28493 137 5 the the DET fcis-28493 137 6 changes change NOUN fcis-28493 137 7 in in ADP fcis-28493 137 8 classification classification NOUN fcis-28493 137 9 accuracy accuracy NOUN fcis-28493 137 10 and and CCONJ fcis-28493 137 11 loss loss NOUN fcis-28493 137 12 value value NOUN fcis-28493 137 13 across across ADP fcis-28493 137 14 five five NUM fcis-28493 137 15 different different ADJ fcis-28493 137 16 algorithms algorithm NOUN fcis-28493 137 17 in in ADP fcis-28493 137 18 the the DET fcis-28493 137 19 task task NOUN fcis-28493 137 20 of of ADP fcis-28493 137 21 grasp grasp NOUN fcis-28493 137 22 detection detection NOUN fcis-28493 137 23 . . PUNCT fcis-28493 138 1 fig fig NOUN fcis-28493 138 2 . . PUNCT fcis-28493 139 1 6 6 NUM fcis-28493 139 2 shows show VERB fcis-28493 139 3 the the DET fcis-28493 139 4 variation variation NOUN fcis-28493 139 5 curves curve NOUN fcis-28493 139 6 of of ADP fcis-28493 139 7 classification classification NOUN fcis-28493 139 8 accuracy accuracy NOUN fcis-28493 139 9 and and CCONJ fcis-28493 139 10 loss loss NOUN fcis-28493 139 11 value value NOUN fcis-28493 139 12 for for ADP fcis-28493 139 13 five five NUM fcis-28493 139 14 algorithms algorithm NOUN fcis-28493 139 15 under under ADP fcis-28493 139 16 a a DET fcis-28493 139 17 single single ADJ fcis-28493 139 18 view view NOUN fcis-28493 139 19 test test NOUN fcis-28493 139 20 . . PUNCT fcis-28493 140 1 the the DET fcis-28493 140 2 blue blue ADJ fcis-28493 140 3 and and CCONJ fcis-28493 140 4 green green ADJ fcis-28493 140 5 curves curve NOUN fcis-28493 140 6 represent represent VERB fcis-28493 140 7 the the DET fcis-28493 140 8 mainstream mainstream NOUN fcis-28493 140 9 pointnetgpd pointnetgpd ADJ fcis-28493 140 10 and and CCONJ fcis-28493 140 11 gpd gpd NUM fcis-28493 140 12 algorithms algorithm NOUN fcis-28493 140 13 , , PUNCT fcis-28493 140 14 respectively respectively ADV fcis-28493 140 15 . . PUNCT fcis-28493 141 1 the the DET fcis-28493 141 2 red red ADJ fcis-28493 141 3 curve curve NOUN fcis-28493 141 4 represents represent VERB fcis-28493 141 5 the the DET fcis-28493 141 6 proposed propose VERB fcis-28493 141 7 point point NOUN fcis-28493 141 8 cloud cloud NOUN fcis-28493 141 9 - - PUNCT fcis-28493 141 10 based base VERB fcis-28493 141 11 grasp grasp NOUN fcis-28493 141 12 detection detection NOUN fcis-28493 141 13 network network NOUN fcis-28493 141 14 ( ( PUNCT fcis-28493 141 15 tes tes NOUN fcis-28493 141 16 - - PUNCT fcis-28493 141 17 net net NOUN fcis-28493 141 18 ) ) PUNCT fcis-28493 141 19 , , PUNCT fcis-28493 141 20 while while SCONJ fcis-28493 141 21 the the DET fcis-28493 141 22 purple purple ADJ fcis-28493 141 23 curve curve NOUN fcis-28493 141 24 shows show VERB fcis-28493 141 25 the the DET fcis-28493 141 26 result result NOUN fcis-28493 141 27 after after ADP fcis-28493 141 28 replacing replace VERB fcis-28493 141 29 the the DET fcis-28493 141 30 graph graph NOUN fcis-28493 141 31 neural neural ADJ fcis-28493 141 32 network network NOUN fcis-28493 141 33 module module NOUN fcis-28493 141 34 in in ADP fcis-28493 141 35 tes tes NOUN fcis-28493 141 36 - - NOUN fcis-28493 141 37 net net NOUN fcis-28493 141 38 with with ADP fcis-28493 141 39 gatv2 gatv2 PROPN fcis-28493 141 40 . . PUNCT fcis-28493 142 1 the the DET fcis-28493 142 2 yellow yellow ADJ fcis-28493 142 3 curve curve NOUN fcis-28493 142 4 represents represent VERB fcis-28493 142 5 the the DET fcis-28493 142 6 original original ADJ fcis-28493 142 7 edgegraspnet edgegraspnet NOUN fcis-28493 142 8 algorithm algorithm NOUN fcis-28493 142 9 . . PUNCT fcis-28493 143 1 the the DET fcis-28493 143 2 left left ADJ fcis-28493 143 3 plot plot NOUN fcis-28493 143 4 shows show VERB fcis-28493 143 5 the the DET fcis-28493 143 6 variation variation NOUN fcis-28493 143 7 of of ADP fcis-28493 143 8 loss loss NOUN fcis-28493 143 9 value value NOUN fcis-28493 143 10 , , PUNCT fcis-28493 143 11 and and CCONJ fcis-28493 143 12 the the DET fcis-28493 143 13 right right ADJ fcis-28493 143 14 plot plot NOUN fcis-28493 143 15 displays display VERB fcis-28493 143 16 the the DET fcis-28493 143 17 trend trend NOUN fcis-28493 143 18 of of ADP fcis-28493 143 19 classification classification NOUN fcis-28493 143 20 accuracy accuracy NOUN fcis-28493 143 21 . . PUNCT fcis-28493 144 1 the the DET fcis-28493 144 2 specific specific ADJ fcis-28493 144 3 experimental experimental ADJ fcis-28493 144 4 data datum NOUN fcis-28493 144 5 is be AUX fcis-28493 144 6 shown show VERB fcis-28493 144 7 in in ADP fcis-28493 144 8 table table NOUN fcis-28493 144 9 1 1 NUM fcis-28493 144 10 . . PUNCT fcis-28493 144 11 table table NOUN fcis-28493 144 12 1 1 NUM fcis-28493 144 13 . . PUNCT fcis-28493 144 14 comparison comparison NOUN fcis-28493 144 15 experiment experiment NOUN fcis-28493 144 16 results result NOUN fcis-28493 144 17 method method VERB fcis-28493 144 18 test_acc% test_acc% NOUN fcis-28493 144 19 loss loss NOUN fcis-28493 144 20 / / SYM fcis-28493 144 21 b b PROPN fcis-28493 144 22 gpd gpd PROPN fcis-28493 144 23 78.6 78.6 NUM fcis-28493 144 24 0.362 0.362 NUM fcis-28493 144 25 pointnetgpd pointnetgpd PROPN fcis-28493 144 26 83.2 83.2 NUM fcis-28493 144 27 0.302 0.302 NUM fcis-28493 144 28 edgegraspnet edgegraspnet NOUN fcis-28493 144 29 88.3 88.3 NUM fcis-28493 144 30 0.282 0.282 NUM fcis-28493 144 31 gatv2conv gatv2conv NUM fcis-28493 144 32 88.9 88.9 NUM fcis-28493 144 33 0.268 0.268 NUM fcis-28493 144 34 tes tes NOUN fcis-28493 144 35 - - PUNCT fcis-28493 144 36 net net NOUN fcis-28493 144 37 89.5 89.5 NUM fcis-28493 144 38 0.262 0.262 NUM fcis-28493 144 39 the the DET fcis-28493 144 40 experimental experimental ADJ fcis-28493 144 41 results result NOUN fcis-28493 144 42 indicate indicate VERB fcis-28493 144 43 significant significant ADJ fcis-28493 144 44 differences difference NOUN fcis-28493 144 45 in in ADP fcis-28493 144 46 classification classification NOUN fcis-28493 144 47 accuracy accuracy NOUN fcis-28493 144 48 and and CCONJ fcis-28493 144 49 loss loss NOUN fcis-28493 144 50 value value NOUN fcis-28493 144 51 among among ADP fcis-28493 144 52 various various ADJ fcis-28493 144 53 grasp grasp NOUN fcis-28493 144 54 detection detection NOUN fcis-28493 144 55 models model NOUN fcis-28493 144 56 . . PUNCT fcis-28493 145 1 according accord VERB fcis-28493 145 2 to to ADP fcis-28493 145 3 the the DET fcis-28493 145 4 data datum NOUN fcis-28493 145 5 in in ADP fcis-28493 145 6 table table NOUN fcis-28493 145 7 1 1 NUM fcis-28493 145 8 , , PUNCT fcis-28493 145 9 tes tes NOUN fcis-28493 145 10 - - ADJ fcis-28493 145 11 net net NOUN fcis-28493 145 12 performs perform VERB fcis-28493 145 13 the the DET fcis-28493 145 14 best good ADJ fcis-28493 145 15 in in ADP fcis-28493 145 16 both both PRON fcis-28493 145 17 classification classification NOUN fcis-28493 145 18 accuracy accuracy NOUN fcis-28493 145 19 and and CCONJ fcis-28493 145 20 loss loss NOUN fcis-28493 145 21 value value NOUN fcis-28493 145 22 , , PUNCT fcis-28493 145 23 achieving achieve VERB fcis-28493 145 24 a a DET fcis-28493 145 25 classification classification NOUN fcis-28493 145 26 accuracy accuracy NOUN fcis-28493 145 27 of of ADP fcis-28493 145 28 89.5 89.5 NUM fcis-28493 145 29 % % NOUN fcis-28493 145 30 and and CCONJ fcis-28493 145 31 a a DET fcis-28493 145 32 minimum minimum ADJ fcis-28493 145 33 loss loss NOUN fcis-28493 145 34 value value NOUN fcis-28493 145 35 of of ADP fcis-28493 145 36 0.262 0.262 NUM fcis-28493 145 37 . . PUNCT fcis-28493 146 1 the the DET fcis-28493 146 2 gatv2conv gatv2conv PROPN fcis-28493 146 3 model model NOUN fcis-28493 146 4 follows follow VERB fcis-28493 146 5 closely closely ADV fcis-28493 146 6 behind behind ADV fcis-28493 146 7 , , PUNCT fcis-28493 146 8 with with ADP fcis-28493 146 9 a a DET fcis-28493 146 10 classification classification NOUN fcis-28493 146 11 accuracy accuracy NOUN fcis-28493 146 12 of of ADP fcis-28493 146 13 88.9 88.9 NUM fcis-28493 146 14 % % NOUN fcis-28493 146 15 and and CCONJ fcis-28493 146 16 a a DET fcis-28493 146 17 loss loss NOUN fcis-28493 146 18 value value NOUN fcis-28493 146 19 of of ADP fcis-28493 146 20 0.268 0.268 NUM fcis-28493 146 21 . . PUNCT fcis-28493 147 1 in in ADP fcis-28493 147 2 contrast contrast NOUN fcis-28493 147 3 , , PUNCT fcis-28493 147 4 pointnetgpd pointnetgpd ADJ fcis-28493 147 5 and and CCONJ fcis-28493 147 6 gpd gpd NUM fcis-28493 147 7 perform perform VERB fcis-28493 147 8 relatively relatively ADV fcis-28493 147 9 worse bad ADJ fcis-28493 147 10 , , PUNCT fcis-28493 147 11 with with ADP fcis-28493 147 12 pointnetgpd pointnetgpd PROPN fcis-28493 147 13 achieving achieve VERB fcis-28493 147 14 a a DET fcis-28493 147 15 classification classification NOUN fcis-28493 147 16 accuracy accuracy NOUN fcis-28493 147 17 of of ADP fcis-28493 147 18 83.2 83.2 NUM fcis-28493 147 19 % % NOUN fcis-28493 147 20 and and CCONJ fcis-28493 147 21 a a DET fcis-28493 147 22 loss loss NOUN fcis-28493 147 23 value value NOUN fcis-28493 147 24 of of ADP fcis-28493 147 25 0.362 0.362 NUM fcis-28493 147 26 , , PUNCT fcis-28493 147 27 while while SCONJ fcis-28493 147 28 gpd gpd PROPN fcis-28493 147 29 's 's PART fcis-28493 147 30 classification classification NOUN fcis-28493 147 31 accuracy accuracy NOUN fcis-28493 147 32 is be AUX fcis-28493 147 33 only only ADV fcis-28493 147 34 78.6 78.6 NUM fcis-28493 147 35 % % NOUN fcis-28493 147 36 , , PUNCT fcis-28493 147 37 with with ADP fcis-28493 147 38 a a DET fcis-28493 147 39 loss loss NOUN fcis-28493 147 40 value value NOUN fcis-28493 147 41 of of ADP fcis-28493 147 42 0.302 0.302 NUM fcis-28493 147 43 . . PUNCT fcis-28493 148 1 combining combine VERB fcis-28493 148 2 the the DET fcis-28493 148 3 training training NOUN fcis-28493 148 4 curves curve NOUN fcis-28493 148 5 from from ADP fcis-28493 148 6 fig fig NOUN fcis-28493 148 7 . . PUNCT fcis-28493 149 1 6 6 NUM fcis-28493 149 2 , , PUNCT fcis-28493 149 3 it it PRON fcis-28493 149 4 can can AUX fcis-28493 149 5 be be AUX fcis-28493 149 6 seen see VERB fcis-28493 149 7 that that SCONJ fcis-28493 149 8 tes tes NOUN fcis-28493 149 9 - - PUNCT fcis-28493 149 10 net net ADJ fcis-28493 149 11 converges converge VERB fcis-28493 149 12 more more ADV fcis-28493 149 13 quickly quickly ADV fcis-28493 149 14 during during ADP fcis-28493 149 15 training training NOUN fcis-28493 149 16 , , PUNCT fcis-28493 149 17 with with ADP fcis-28493 149 18 the the DET fcis-28493 149 19 loss loss NOUN fcis-28493 149 20 value value NOUN fcis-28493 149 21 dropping drop VERB fcis-28493 149 22 rapidly rapidly ADV fcis-28493 149 23 . . PUNCT fcis-28493 150 1 it it PRON fcis-28493 150 2 maintains maintain VERB fcis-28493 150 3 a a DET fcis-28493 150 4 low low ADJ fcis-28493 150 5 loss loss NOUN fcis-28493 150 6 value value NOUN fcis-28493 150 7 and and CCONJ fcis-28493 150 8 high high ADJ fcis-28493 150 9 classification classification NOUN fcis-28493 150 10 accuracy accuracy NOUN fcis-28493 150 11 throughout throughout ADP fcis-28493 150 12 the the DET fcis-28493 150 13 entire entire ADJ fcis-28493 150 14 training training NOUN fcis-28493 150 15 phase phase NOUN fcis-28493 150 16 , , PUNCT fcis-28493 150 17 demonstrating demonstrate VERB fcis-28493 150 18 that that SCONJ fcis-28493 150 19 tes tes NOUN fcis-28493 150 20 - - PUNCT fcis-28493 150 21 net net NOUN fcis-28493 150 22 has have VERB fcis-28493 150 23 high high ADJ fcis-28493 150 24 robustness robustness NOUN fcis-28493 150 25 and and CCONJ fcis-28493 150 26 effectiveness effectiveness NOUN fcis-28493 150 27 in in ADP fcis-28493 150 28 grasp grasp NOUN fcis-28493 150 29 detection detection NOUN fcis-28493 150 30 tasks task NOUN fcis-28493 150 31 . . PUNCT fcis-28493 151 1 overall overall ADJ fcis-28493 151 2 , , PUNCT fcis-28493 151 3 tes tes NOUN fcis-28493 151 4 - - ADJ fcis-28493 151 5 net net NOUN fcis-28493 151 6 shows show VERB fcis-28493 151 7 the the DET fcis-28493 151 8 best good ADJ fcis-28493 151 9 classification classification NOUN fcis-28493 151 10 performance performance NOUN fcis-28493 151 11 , , PUNCT fcis-28493 151 12 outperforming outperform VERB fcis-28493 151 13 existing exist VERB fcis-28493 151 14 mainstream mainstream NOUN fcis-28493 151 15 grasp grasp NOUN fcis-28493 151 16 detection detection NOUN fcis-28493 151 17 models model NOUN fcis-28493 151 18 , , PUNCT fcis-28493 151 19 validating validate VERB fcis-28493 151 20 its its PRON fcis-28493 151 21 ability ability NOUN fcis-28493 151 22 to to PART fcis-28493 151 23 handle handle VERB fcis-28493 151 24 point point NOUN fcis-28493 151 25 cloud cloud ADJ fcis-28493 151 26 data datum NOUN fcis-28493 151 27 in in ADP fcis-28493 151 28 grasp grasp NOUN fcis-28493 151 29 detection detection NOUN fcis-28493 151 30 tasks task NOUN fcis-28493 151 31 . . PUNCT fcis-28493 152 1 6 6 X fcis-28493 152 2 . . X fcis-28493 152 3 simulation simulation NOUN fcis-28493 152 4 validation validation NOUN fcis-28493 152 5 the the DET fcis-28493 152 6 simulation simulation NOUN fcis-28493 152 7 in in ADP fcis-28493 152 8 this this DET fcis-28493 152 9 paper paper NOUN fcis-28493 152 10 was be AUX fcis-28493 152 11 conducted conduct VERB fcis-28493 152 12 in in ADP fcis-28493 152 13 the the DET fcis-28493 152 14 pybullet pybullet NOUN fcis-28493 152 15 simulation simulation NOUN fcis-28493 152 16 environment environment NOUN fcis-28493 152 17 using use VERB fcis-28493 152 18 the the DET fcis-28493 152 19 grasping grasp VERB fcis-28493 152 20 simulator simulator NOUN fcis-28493 152 21 developed develop VERB fcis-28493 152 22 by by ADP fcis-28493 152 23 breyer breyer NOUN fcis-28493 152 24 et et PROPN fcis-28493 152 25 al al PROPN fcis-28493 152 26 . . PUNCT fcis-28493 153 1 [ [ X fcis-28493 153 2 13 13 NUM fcis-28493 153 3 ] ] PUNCT fcis-28493 153 4 , , PUNCT fcis-28493 153 5 which which PRON fcis-28493 153 6 includes include VERB fcis-28493 153 7 the the DET fcis-28493 153 8 gripper gripper NOUN fcis-28493 153 9 of of ADP fcis-28493 153 10 the the DET fcis-28493 153 11 franka franka PROPN fcis-28493 153 12 robot robot PROPN fcis-28493 153 13 arm arm PROPN fcis-28493 153 14 . . PUNCT fcis-28493 154 1 the the DET fcis-28493 154 2 grasping grasp VERB fcis-28493 154 3 simulator simulator NOUN fcis-28493 154 4 contains contain VERB fcis-28493 154 5 two two NUM fcis-28493 154 6 different different ADJ fcis-28493 154 7 grasping grasp VERB fcis-28493 154 8 environments environment NOUN fcis-28493 154 9 : : PUNCT fcis-28493 154 10 one one NUM fcis-28493 154 11 is be AUX fcis-28493 154 12 a a DET fcis-28493 154 13 vertically vertically ADV fcis-28493 154 14 placed place VERB fcis-28493 154 15 environment environment NOUN fcis-28493 154 16 , , PUNCT fcis-28493 154 17 and and CCONJ fcis-28493 154 18 the the DET fcis-28493 154 19 other other ADJ fcis-28493 154 20 is be AUX fcis-28493 154 21 a a DET fcis-28493 154 22 randomly randomly ADV fcis-28493 154 23 placed place VERB fcis-28493 154 24 environment environment NOUN fcis-28493 154 25 . . PUNCT fcis-28493 155 1 the the DET fcis-28493 155 2 simulation simulation NOUN fcis-28493 155 3 environment environment NOUN fcis-28493 155 4 is be AUX fcis-28493 155 5 shown show VERB fcis-28493 155 6 in in ADP fcis-28493 155 7 fig fig NOUN fcis-28493 155 8 . . PUNCT fcis-28493 156 1 7 7 X fcis-28493 156 2 . . X fcis-28493 156 3 fig fig NOUN fcis-28493 156 4 7 7 NUM fcis-28493 156 5 . . PUNCT fcis-28493 157 1 left leave VERB fcis-28493 157 2 : : PUNCT fcis-28493 157 3 vertically vertically ADV fcis-28493 157 4 placed place VERB fcis-28493 157 5 , , PUNCT fcis-28493 157 6 right right ADV fcis-28493 157 7 : : PUNCT fcis-28493 157 8 randomly randomly ADV fcis-28493 157 9 placed place VERB fcis-28493 157 10 at at ADP fcis-28493 157 11 the the DET fcis-28493 157 12 beginning beginning NOUN fcis-28493 157 13 of of ADP fcis-28493 157 14 each each DET fcis-28493 157 15 test test NOUN fcis-28493 157 16 round round NOUN fcis-28493 157 17 , , PUNCT fcis-28493 157 18 five five NUM fcis-28493 157 19 random random ADJ fcis-28493 157 20 objects object NOUN fcis-28493 157 21 are be AUX fcis-28493 157 22 placed place VERB fcis-28493 157 23 in in ADP fcis-28493 157 24 the the DET fcis-28493 157 25 workspace workspace NOUN fcis-28493 157 26 , , PUNCT fcis-28493 157 27 as as SCONJ fcis-28493 157 28 shown show VERB fcis-28493 157 29 in in ADP fcis-28493 157 30 fig fig NOUN fcis-28493 157 31 . . PUNCT fcis-28493 158 1 8 8 NUM fcis-28493 158 2 . . PUNCT fcis-28493 159 1 after after SCONJ fcis-28493 159 2 the the DET fcis-28493 159 3 objects object NOUN fcis-28493 159 4 are be AUX fcis-28493 159 5 loaded load VERB fcis-28493 159 6 , , PUNCT fcis-28493 159 7 grasping grasp VERB fcis-28493 159 8 is be AUX fcis-28493 159 9 performed perform VERB fcis-28493 159 10 sequentially sequentially ADV fcis-28493 159 11 . . PUNCT fcis-28493 160 1 during during ADP fcis-28493 160 2 the the DET fcis-28493 160 3 grasping grasp VERB fcis-28493 160 4 process process NOUN fcis-28493 160 5 , , PUNCT fcis-28493 160 6 the the DET fcis-28493 160 7 rgb rgb PROPN fcis-28493 160 8 data datum NOUN fcis-28493 160 9 of of ADP fcis-28493 160 10 the the DET fcis-28493 160 11 objects object NOUN fcis-28493 160 12 and and CCONJ fcis-28493 160 13 the the DET fcis-28493 160 14 segmentation segmentation NOUN fcis-28493 160 15 results result NOUN fcis-28493 160 16 can can AUX fcis-28493 160 17 be be AUX fcis-28493 160 18 observed observe VERB fcis-28493 160 19 from from ADP fcis-28493 160 20 the the DET fcis-28493 160 21 workspace workspace NOUN fcis-28493 160 22 interface interface NOUN fcis-28493 160 23 . . PUNCT fcis-28493 161 1 before before ADP fcis-28493 161 2 each each DET fcis-28493 161 3 grasp grasp NOUN fcis-28493 161 4 attempt attempt NOUN fcis-28493 161 5 , , PUNCT fcis-28493 161 6 a a DET fcis-28493 161 7 depth depth NOUN fcis-28493 161 8 image image NOUN fcis-28493 161 9 of of ADP fcis-28493 161 10 the the DET fcis-28493 161 11 scene scene NOUN fcis-28493 161 12 is be AUX fcis-28493 161 13 captured capture VERB fcis-28493 161 14 , , PUNCT fcis-28493 161 15 gaussian gaussian ADJ fcis-28493 161 16 noise noise NOUN fcis-28493 161 17 is be AUX fcis-28493 161 18 added add VERB fcis-28493 161 19 , , PUNCT fcis-28493 161 20 and and CCONJ fcis-28493 161 21 the the DET fcis-28493 161 22 point point NOUN fcis-28493 161 23 cloud cloud NOUN fcis-28493 161 24 or or CCONJ fcis-28493 161 25 tsdf tsdf NOUN fcis-28493 161 26 is be AUX fcis-28493 161 27 extracted extract VERB fcis-28493 161 28 and and CCONJ fcis-28493 161 29 input input NOUN fcis-28493 161 30 into into ADP fcis-28493 161 31 the the DET fcis-28493 161 32 model model NOUN fcis-28493 161 33 . . PUNCT fcis-28493 162 1 after after SCONJ fcis-28493 162 2 the the DET fcis-28493 162 3 model model NOUN fcis-28493 162 4 outputs output VERB fcis-28493 162 5 the the DET fcis-28493 162 6 grasp grasp NOUN fcis-28493 162 7 score score NOUN fcis-28493 162 8 , , PUNCT fcis-28493 162 9 the the DET fcis-28493 162 10 grasping grasp VERB fcis-28493 162 11 plan plan NOUN fcis-28493 162 12 with with ADP fcis-28493 162 13 the the DET fcis-28493 162 14 highest high ADJ fcis-28493 162 15 score score NOUN fcis-28493 162 16 is be AUX fcis-28493 162 17 executed execute VERB fcis-28493 162 18 . . PUNCT fcis-28493 163 1 upon upon SCONJ fcis-28493 163 2 successful successful ADJ fcis-28493 163 3 grasping grasping NOUN fcis-28493 163 4 , , PUNCT fcis-28493 163 5 the the DET fcis-28493 163 6 object object NOUN fcis-28493 163 7 is be AUX fcis-28493 163 8 removed remove VERB fcis-28493 163 9 . . PUNCT fcis-28493 164 1 the the DET fcis-28493 164 2 test test NOUN fcis-28493 164 3 ends end VERB fcis-28493 164 4 when when SCONJ fcis-28493 164 5 all all DET fcis-28493 164 6 objects object NOUN fcis-28493 164 7 are be AUX fcis-28493 164 8 cleared clear VERB fcis-28493 164 9 or or CCONJ fcis-28493 164 10 when when SCONJ fcis-28493 164 11 two two NUM fcis-28493 164 12 consecutive consecutive ADJ fcis-28493 164 13 grasp grasp NOUN fcis-28493 164 14 attempts attempt NOUN fcis-28493 164 15 fail fail VERB fcis-28493 164 16 . . PUNCT fcis-28493 165 1 a a DET fcis-28493 165 2 total total NOUN fcis-28493 165 3 of of ADP fcis-28493 165 4 100 100 NUM fcis-28493 165 5 test test NOUN fcis-28493 165 6 rounds round NOUN fcis-28493 165 7 are be AUX fcis-28493 165 8 performed perform VERB fcis-28493 165 9 . . PUNCT fcis-28493 166 1 fig fig NOUN fcis-28493 166 2 8 8 NUM fcis-28493 166 3 . . PUNCT fcis-28493 167 1 workspace workspace NOUN fcis-28493 167 2 environment environment NOUN fcis-28493 167 3 to to PART fcis-28493 167 4 objectively objectively ADV fcis-28493 167 5 evaluate evaluate VERB fcis-28493 167 6 the the DET fcis-28493 167 7 performance performance NOUN fcis-28493 167 8 of of ADP fcis-28493 167 9 the the DET fcis-28493 167 10 improved improved ADJ fcis-28493 167 11 network network NOUN fcis-28493 167 12 , , PUNCT fcis-28493 167 13 this this DET fcis-28493 167 14 paper paper NOUN fcis-28493 167 15 uses use VERB fcis-28493 167 16 gsr gsr NOUN fcis-28493 167 17 ( ( PUNCT fcis-28493 167 18 grasp grasp VERB fcis-28493 167 19 success success NOUN fcis-28493 167 20 rate rate NOUN fcis-28493 167 21 ) ) PUNCT fcis-28493 167 22 and and CCONJ fcis-28493 167 23 dr dr PROPN fcis-28493 167 24 ( ( PUNCT fcis-28493 167 25 clearance clearance NOUN fcis-28493 167 26 rate rate NOUN fcis-28493 167 27 ) ) PUNCT fcis-28493 167 28 as as ADP fcis-28493 167 29 simulation simulation NOUN fcis-28493 167 30 evaluation evaluation NOUN fcis-28493 167 31 metrics metric NOUN fcis-28493 167 32 , , PUNCT fcis-28493 167 33 with with ADP fcis-28493 167 34 the the DET fcis-28493 167 35 specific specific ADJ fcis-28493 167 36 calculation calculation NOUN fcis-28493 167 37 formulas formula NOUN fcis-28493 167 38 shown show VERB fcis-28493 167 39 in in ADP fcis-28493 167 40 equation equation NOUN fcis-28493 167 41 13 13 NUM fcis-28493 167 42 - - SYM fcis-28493 167 43 14 14 NUM fcis-28493 167 44 . . PUNCT fcis-28493 168 1 ( ( PUNCT fcis-28493 168 2 13 13 NUM fcis-28493 168 3 ) ) PUNCT fcis-28493 168 4 ( ( PUNCT fcis-28493 168 5 14 14 NUM fcis-28493 168 6 ) ) PUNCT fcis-28493 168 7 in in ADP fcis-28493 168 8 the the DET fcis-28493 168 9 pybullet pybullet NOUN fcis-28493 168 10 simulation simulation NOUN fcis-28493 168 11 environment environment NOUN fcis-28493 168 12 , , PUNCT fcis-28493 168 13 this this DET fcis-28493 168 14 paper paper NOUN fcis-28493 168 15 70 70 NUM fcis-28493 168 16 conducted conduct VERB fcis-28493 168 17 simulation simulation NOUN fcis-28493 168 18 comparison comparison NOUN fcis-28493 168 19 experiments experiment NOUN fcis-28493 168 20 with with ADP fcis-28493 168 21 four four NUM fcis-28493 168 22 algorithms algorithm NOUN fcis-28493 168 23 : : PUNCT fcis-28493 168 24 pointnetgpd pointnetgpd PROPN fcis-28493 168 25 , , PUNCT fcis-28493 168 26 gpd gpd PROPN fcis-28493 168 27 , , PUNCT fcis-28493 168 28 edgegrasp edgegrasp NOUN fcis-28493 168 29 - - PUNCT fcis-28493 168 30 net net NOUN fcis-28493 168 31 , , PUNCT fcis-28493 168 32 and and CCONJ fcis-28493 168 33 tesnet tesnet NOUN fcis-28493 168 34 . . PUNCT fcis-28493 169 1 the the DET fcis-28493 169 2 specific specific ADJ fcis-28493 169 3 simulation simulation NOUN fcis-28493 169 4 grasping grasp VERB fcis-28493 169 5 experiment experiment NOUN fcis-28493 169 6 data datum NOUN fcis-28493 169 7 is be AUX fcis-28493 169 8 shown show VERB fcis-28493 169 9 in in ADP fcis-28493 169 10 table table NOUN fcis-28493 169 11 2 2 NUM fcis-28493 169 12 . . PUNCT fcis-28493 169 13 table table NOUN fcis-28493 169 14 2 2 NUM fcis-28493 169 15 . . PUNCT fcis-28493 169 16 grasp grasp VERB fcis-28493 169 17 success success NOUN fcis-28493 169 18 rate rate NOUN fcis-28493 169 19 in in ADP fcis-28493 169 20 different different ADJ fcis-28493 169 21 environments environment NOUN fcis-28493 169 22 method method VERB fcis-28493 169 23 vertically vertically ADV fcis-28493 169 24 gsr(% gsr(% NOUN fcis-28493 169 25 ) ) PUNCT fcis-28493 169 26 dr(% dr(% X fcis-28493 169 27 ) ) PUNCT fcis-28493 169 28 randomly randomly ADV fcis-28493 169 29 gsr(% gsr(% NOUN fcis-28493 169 30 ) ) PUNCT fcis-28493 169 31 dr(% dr(% PUNCT fcis-28493 169 32 ) ) PUNCT fcis-28493 170 1 gpd gpd PROPN fcis-28493 170 2 78.5 78.5 NUM fcis-28493 170 3 81.3 81.3 NUM fcis-28493 170 4 75.6 75.6 NUM fcis-28493 170 5 77.2 77.2 NUM fcis-28493 170 6 pointnetgpd pointnetgpd ADP fcis-28493 170 7 81.1 81.1 NUM fcis-28493 170 8 85.4 85.4 NUM fcis-28493 170 9 77.9 77.9 NUM fcis-28493 170 10 79.8 79.8 NUM fcis-28493 170 11 edgegrasp edgegrasp NOUN fcis-28493 170 12 - - PUNCT fcis-28493 170 13 net net NOUN fcis-28493 170 14 90.6 90.6 NUM fcis-28493 170 15 92.2 92.2 NUM fcis-28493 170 16 87.1 87.1 NUM fcis-28493 170 17 90.2 90.2 NUM fcis-28493 170 18 tes tes NOUN fcis-28493 170 19 - - PUNCT fcis-28493 170 20 net net ADJ fcis-28493 170 21 94.1 94.1 NUM fcis-28493 170 22 95.2 95.2 NUM fcis-28493 170 23 92.2 92.2 NUM fcis-28493 170 24 93.5 93.5 NUM fcis-28493 170 25 the the DET fcis-28493 170 26 results result NOUN fcis-28493 170 27 in in ADP fcis-28493 170 28 table table NOUN fcis-28493 170 29 2 2 NUM fcis-28493 170 30 show show NOUN fcis-28493 170 31 that that SCONJ fcis-28493 170 32 tes tes NOUN fcis-28493 170 33 - - PUNCT fcis-28493 170 34 net net ADJ fcis-28493 170 35 excels excel NOUN fcis-28493 170 36 in in ADP fcis-28493 170 37 both both DET fcis-28493 170 38 grasp grasp NOUN fcis-28493 170 39 success success NOUN fcis-28493 170 40 rate rate NOUN fcis-28493 170 41 ( ( PUNCT fcis-28493 170 42 gsr gsr NOUN fcis-28493 170 43 ) ) PUNCT fcis-28493 170 44 and and CCONJ fcis-28493 170 45 detection detection NOUN fcis-28493 170 46 rate rate NOUN fcis-28493 170 47 ( ( PUNCT fcis-28493 170 48 dr dr PROPN fcis-28493 170 49 ) ) PUNCT fcis-28493 170 50 . . PUNCT fcis-28493 171 1 in in ADP fcis-28493 171 2 the the DET fcis-28493 171 3 vertical vertical ADJ fcis-28493 171 4 environment environment NOUN fcis-28493 171 5 , , PUNCT fcis-28493 171 6 tes tes NOUN fcis-28493 171 7 - - PUNCT fcis-28493 171 8 net net NOUN fcis-28493 171 9 achieves achieve VERB fcis-28493 171 10 a a DET fcis-28493 171 11 gsr gsr NOUN fcis-28493 171 12 of of ADP fcis-28493 171 13 94.1 94.1 NUM fcis-28493 171 14 % % NOUN fcis-28493 171 15 and and CCONJ fcis-28493 171 16 a a DET fcis-28493 171 17 dr dr NOUN fcis-28493 171 18 of of ADP fcis-28493 171 19 95.2 95.2 NUM fcis-28493 171 20 % % NOUN fcis-28493 171 21 . . PUNCT fcis-28493 172 1 in in ADP fcis-28493 172 2 the the DET fcis-28493 172 3 random random ADJ fcis-28493 172 4 environment environment NOUN fcis-28493 172 5 , , PUNCT fcis-28493 172 6 the the DET fcis-28493 172 7 gsr gsr NOUN fcis-28493 172 8 and and CCONJ fcis-28493 172 9 dr dr PROPN fcis-28493 172 10 are be AUX fcis-28493 172 11 92.2 92.2 NUM fcis-28493 172 12 % % NOUN fcis-28493 172 13 and and CCONJ fcis-28493 172 14 93.5 93.5 NUM fcis-28493 172 15 % % NOUN fcis-28493 172 16 , , PUNCT fcis-28493 172 17 respectively respectively ADV fcis-28493 172 18 . . PUNCT fcis-28493 173 1 these these DET fcis-28493 173 2 metrics metric NOUN fcis-28493 173 3 significantly significantly ADV fcis-28493 173 4 outperform outperform VERB fcis-28493 173 5 other other ADJ fcis-28493 173 6 methods method NOUN fcis-28493 173 7 . . PUNCT fcis-28493 174 1 this this PRON fcis-28493 174 2 indicates indicate VERB fcis-28493 174 3 that that SCONJ fcis-28493 174 4 tes tes NOUN fcis-28493 174 5 - - PUNCT fcis-28493 174 6 net net NOUN fcis-28493 174 7 provides provide VERB fcis-28493 174 8 higher high ADJ fcis-28493 174 9 grasp grasp NOUN fcis-28493 174 10 success success NOUN fcis-28493 174 11 and and CCONJ fcis-28493 174 12 detection detection NOUN fcis-28493 174 13 rates rate NOUN fcis-28493 174 14 across across ADP fcis-28493 174 15 different different ADJ fcis-28493 174 16 scenarios scenario NOUN fcis-28493 174 17 , , PUNCT fcis-28493 174 18 demonstrating demonstrate VERB fcis-28493 174 19 its its PRON fcis-28493 174 20 superiority superiority NOUN fcis-28493 174 21 in in ADP fcis-28493 174 22 handling handle VERB fcis-28493 174 23 grasping grasp VERB fcis-28493 174 24 tasks task NOUN fcis-28493 174 25 . . PUNCT fcis-28493 175 1 7 7 X fcis-28493 175 2 . . X fcis-28493 175 3 summary summary NOUN fcis-28493 175 4 to to PART fcis-28493 175 5 address address VERB fcis-28493 175 6 the the DET fcis-28493 175 7 issue issue NOUN fcis-28493 175 8 of of ADP fcis-28493 175 9 low low ADJ fcis-28493 175 10 object object NOUN fcis-28493 175 11 grasping grasp VERB fcis-28493 175 12 accuracy accuracy NOUN fcis-28493 175 13 in in ADP fcis-28493 175 14 robotic robotic ADJ fcis-28493 175 15 arm arm NOUN fcis-28493 175 16 grasp grasp NOUN fcis-28493 175 17 detection detection NOUN fcis-28493 175 18 , , PUNCT fcis-28493 175 19 this this DET fcis-28493 175 20 paper paper NOUN fcis-28493 175 21 proposes propose VERB fcis-28493 175 22 an an DET fcis-28493 175 23 improved improved ADJ fcis-28493 175 24 point point NOUN fcis-28493 175 25 cloud cloud NOUN fcis-28493 175 26 - - PUNCT fcis-28493 175 27 based base VERB fcis-28493 175 28 grasp grasp NOUN fcis-28493 175 29 detection detection NOUN fcis-28493 175 30 network network NOUN fcis-28493 175 31 , , PUNCT fcis-28493 175 32 tes tes NOUN fcis-28493 175 33 - - NOUN fcis-28493 175 34 net net NOUN fcis-28493 175 35 , , PUNCT fcis-28493 175 36 based base VERB fcis-28493 175 37 on on ADP fcis-28493 175 38 the the DET fcis-28493 175 39 edge edge NOUN fcis-28493 175 40 grasp grasp NOUN fcis-28493 175 41 detection detection NOUN fcis-28493 175 42 network network NOUN fcis-28493 175 43 . . PUNCT fcis-28493 176 1 by by ADP fcis-28493 176 2 incorporating incorporate VERB fcis-28493 176 3 the the DET fcis-28493 176 4 pointtransformerconv pointtransformerconv ADJ fcis-28493 176 5 and and CCONJ fcis-28493 176 6 sageconv sageconv NOUN fcis-28493 176 7 graph graph NOUN fcis-28493 176 8 neural neural ADJ fcis-28493 176 9 modules module NOUN fcis-28493 176 10 , , PUNCT fcis-28493 176 11 the the DET fcis-28493 176 12 network network NOUN fcis-28493 176 13 enhances enhance VERB fcis-28493 176 14 the the DET fcis-28493 176 15 ability ability NOUN fcis-28493 176 16 to to PART fcis-28493 176 17 capture capture VERB fcis-28493 176 18 global global ADJ fcis-28493 176 19 information information NOUN fcis-28493 176 20 , , PUNCT fcis-28493 176 21 thereby thereby ADV fcis-28493 176 22 improving improve VERB fcis-28493 176 23 feature feature NOUN fcis-28493 176 24 extraction extraction NOUN fcis-28493 176 25 and and CCONJ fcis-28493 176 26 representation representation NOUN fcis-28493 176 27 capabilities capability NOUN fcis-28493 176 28 . . PUNCT fcis-28493 177 1 this this DET fcis-28493 177 2 approach approach NOUN fcis-28493 177 3 solves solve VERB fcis-28493 177 4 the the DET fcis-28493 177 5 problem problem NOUN fcis-28493 177 6 of of ADP fcis-28493 177 7 neglecting neglect VERB fcis-28493 177 8 the the DET fcis-28493 177 9 relationships relationship NOUN fcis-28493 177 10 between between ADP fcis-28493 177 11 points point NOUN fcis-28493 177 12 during during ADP fcis-28493 177 13 feature feature NOUN fcis-28493 177 14 extraction extraction NOUN fcis-28493 177 15 and and CCONJ fcis-28493 177 16 improves improve VERB fcis-28493 177 17 overall overall ADJ fcis-28493 177 18 network network NOUN fcis-28493 177 19 performance performance NOUN fcis-28493 177 20 . . PUNCT fcis-28493 178 1 the the DET fcis-28493 178 2 model model NOUN fcis-28493 178 3 was be AUX fcis-28493 178 4 trained train VERB fcis-28493 178 5 and and CCONJ fcis-28493 178 6 validated validate VERB fcis-28493 178 7 on on ADP fcis-28493 178 8 public public ADJ fcis-28493 178 9 datasets dataset NOUN fcis-28493 178 10 such such ADJ fcis-28493 178 11 as as ADP fcis-28493 178 12 ycb ycb NOUN fcis-28493 178 13 , , PUNCT fcis-28493 178 14 bigbird bigbird PROPN fcis-28493 178 15 , , PUNCT fcis-28493 178 16 and and CCONJ fcis-28493 178 17 kit kit NOUN fcis-28493 178 18 , , PUNCT fcis-28493 178 19 and and CCONJ fcis-28493 178 20 simulation simulation NOUN fcis-28493 178 21 grasping grasp VERB fcis-28493 178 22 experiments experiment NOUN fcis-28493 178 23 were be AUX fcis-28493 178 24 conducted conduct VERB fcis-28493 178 25 in in ADP fcis-28493 178 26 two two NUM fcis-28493 178 27 different different ADJ fcis-28493 178 28 grasping grasp VERB fcis-28493 178 29 environments environment NOUN fcis-28493 178 30 in in ADP fcis-28493 178 31 pybullet pybullet NOUN fcis-28493 178 32 . . PUNCT fcis-28493 179 1 experimental experimental ADJ fcis-28493 179 2 results result NOUN fcis-28493 179 3 show show VERB fcis-28493 179 4 that that SCONJ fcis-28493 179 5 tes tes NOUN fcis-28493 179 6 - - PUNCT fcis-28493 179 7 net net NOUN fcis-28493 179 8 achieves achieve VERB fcis-28493 179 9 a a DET fcis-28493 179 10 classification classification NOUN fcis-28493 179 11 accuracy accuracy NOUN fcis-28493 179 12 of of ADP fcis-28493 179 13 89.5 89.5 NUM fcis-28493 179 14 % % NOUN fcis-28493 179 15 , , PUNCT fcis-28493 179 16 a a DET fcis-28493 179 17 loss loss NOUN fcis-28493 179 18 value value NOUN fcis-28493 179 19 of of ADP fcis-28493 179 20 0.262 0.262 NUM fcis-28493 179 21 , , PUNCT fcis-28493 179 22 a a DET fcis-28493 179 23 success success NOUN fcis-28493 179 24 rate rate NOUN fcis-28493 179 25 of of ADP fcis-28493 179 26 94.1 94.1 NUM fcis-28493 179 27 % % NOUN fcis-28493 179 28 in in ADP fcis-28493 179 29 the the DET fcis-28493 179 30 vertical vertical ADJ fcis-28493 179 31 grasping grasp VERB fcis-28493 179 32 environment environment NOUN fcis-28493 179 33 , , PUNCT fcis-28493 179 34 and and CCONJ fcis-28493 179 35 a a DET fcis-28493 179 36 success success NOUN fcis-28493 179 37 rate rate NOUN fcis-28493 179 38 of of ADP fcis-28493 179 39 92.2 92.2 NUM fcis-28493 179 40 % % NOUN fcis-28493 179 41 in in ADP fcis-28493 179 42 the the DET fcis-28493 179 43 random random ADJ fcis-28493 179 44 grasping grasp VERB fcis-28493 179 45 environment environment NOUN fcis-28493 179 46 . . PUNCT fcis-28493 180 1 compared compare VERB fcis-28493 180 2 to to ADP fcis-28493 180 3 existing exist VERB fcis-28493 180 4 mainstream mainstream NOUN fcis-28493 180 5 point point NOUN fcis-28493 180 6 cloud cloud NOUN fcis-28493 180 7 grasp grasp NOUN fcis-28493 180 8 detection detection NOUN fcis-28493 180 9 methods method NOUN fcis-28493 180 10 such such ADJ fcis-28493 180 11 as as ADP fcis-28493 180 12 pointnetgpd pointnetgpd PROPN fcis-28493 180 13 and and CCONJ fcis-28493 180 14 gpd gpd NOUN fcis-28493 180 15 , , PUNCT fcis-28493 180 16 tes tes NOUN fcis-28493 180 17 - - ADJ fcis-28493 180 18 net net NOUN fcis-28493 180 19 demonstrates demonstrate VERB fcis-28493 180 20 superior superior ADJ fcis-28493 180 21 performance performance NOUN fcis-28493 180 22 in in ADP fcis-28493 180 23 terms term NOUN fcis-28493 180 24 of of ADP fcis-28493 180 25 model model NOUN fcis-28493 180 26 generalization generalization NOUN fcis-28493 180 27 ability ability NOUN fcis-28493 180 28 , , PUNCT fcis-28493 180 29 robustness robustness NOUN fcis-28493 180 30 , , PUNCT fcis-28493 180 31 and and CCONJ fcis-28493 180 32 stability stability NOUN fcis-28493 180 33 . . PUNCT fcis-28493 181 1 acknowledgments acknowledgment NOUN fcis-28493 181 2 jilin jilin PROPN fcis-28493 181 3 province province PROPN fcis-28493 181 4 science science PROPN fcis-28493 181 5 and and CCONJ fcis-28493 181 6 technology technology NOUN fcis-28493 181 7 development development NOUN fcis-28493 181 8 item item NOUN fcis-28493 181 9 ( ( PUNCT fcis-28493 181 10 ydzj202201zyts555 ydzj202201zyts555 NOUN fcis-28493 181 11 ) ) PUNCT fcis-28493 181 12 . . PUNCT fcis-28493 182 1 references reference NOUN fcis-28493 182 2 [ [ X fcis-28493 182 3 1 1 X fcis-28493 182 4 ] ] X fcis-28493 182 5 liu liu PROPN fcis-28493 182 6 j j PROPN fcis-28493 182 7 , , PUNCT fcis-28493 182 8 sun sun PROPN fcis-28493 182 9 w w PROPN fcis-28493 182 10 , , PUNCT fcis-28493 182 11 yang yang PROPN fcis-28493 182 12 h h PROPN fcis-28493 182 13 , , PUNCT fcis-28493 182 14 et et PROPN fcis-28493 182 15 al al PROPN fcis-28493 182 16 . . PUNCT fcis-28493 183 1 deep deep ADJ fcis-28493 183 2 learning learning NOUN fcis-28493 183 3 - - PUNCT fcis-28493 183 4 based base VERB fcis-28493 183 5 object object NOUN fcis-28493 183 6 pose pose VERB fcis-28493 183 7 estimation estimation NOUN fcis-28493 183 8 : : PUNCT fcis-28493 183 9 a a DET fcis-28493 183 10 comprehensive comprehensive ADJ fcis-28493 183 11 survey[j survey[j NOUN fcis-28493 183 12 ] ] PUNCT fcis-28493 183 13 . . PUNCT fcis-28493 184 1 arxiv arxiv PROPN fcis-28493 184 2 preprint preprint PROPN fcis-28493 184 3 arxiv arxiv PROPN fcis-28493 184 4 : : PUNCT fcis-28493 184 5 2405 2405 NUM fcis-28493 184 6 . . PUNCT fcis-28493 185 1 07801 07801 NUM fcis-28493 185 2 , , PUNCT fcis-28493 185 3 2024 2024 NUM fcis-28493 185 4 . . PUNCT fcis-28493 186 1 [ [ X fcis-28493 186 2 2 2 NUM fcis-28493 186 3 ] ] PUNCT fcis-28493 186 4 ten ten NUM fcis-28493 186 5 pas pas NOUN fcis-28493 186 6 a a PRON fcis-28493 186 7 , , PUNCT fcis-28493 186 8 gualtieri gualtieri PROPN fcis-28493 186 9 m m PROPN fcis-28493 186 10 , , PUNCT fcis-28493 186 11 saenko saenko PROPN fcis-28493 186 12 k k NOUN fcis-28493 186 13 , , PUNCT fcis-28493 186 14 et et PROPN fcis-28493 186 15 al al PROPN fcis-28493 186 16 . . PROPN fcis-28493 186 17 grasp grasp NOUN fcis-28493 186 18 pose pose NOUN fcis-28493 186 19 detection detection NOUN fcis-28493 186 20 in in ADP fcis-28493 186 21 point point NOUN fcis-28493 186 22 clouds[j clouds[j PROPN fcis-28493 186 23 ] ] X fcis-28493 186 24 . . PUNCT fcis-28493 187 1 the the DET fcis-28493 187 2 international international ADJ fcis-28493 187 3 journal journal NOUN fcis-28493 187 4 of of ADP fcis-28493 187 5 robotics robotic NOUN fcis-28493 187 6 research research NOUN fcis-28493 187 7 , , PUNCT fcis-28493 187 8 2017 2017 NUM fcis-28493 187 9 , , PUNCT fcis-28493 187 10 36(13 36(13 NUM fcis-28493 187 11 - - SYM fcis-28493 187 12 14 14 NUM fcis-28493 187 13 ): ): PUNCT fcis-28493 187 14 1455 1455 NUM fcis-28493 187 15 - - SYM fcis-28493 187 16 1473 1473 NUM fcis-28493 187 17 . . PUNCT fcis-28493 188 1 [ [ X fcis-28493 188 2 3 3 X fcis-28493 188 3 ] ] X fcis-28493 188 4 liang liang PROPN fcis-28493 188 5 h h PROPN fcis-28493 188 6 , , PUNCT fcis-28493 188 7 ma ma PROPN fcis-28493 188 8 x x PROPN fcis-28493 188 9 , , PUNCT fcis-28493 188 10 li li PROPN fcis-28493 188 11 s s PROPN fcis-28493 188 12 , , PUNCT fcis-28493 188 13 et et PROPN fcis-28493 188 14 al al PROPN fcis-28493 188 15 . . PROPN fcis-28493 188 16 pointnetgpd pointnetgpd PROPN fcis-28493 188 17 : : PUNCT fcis-28493 188 18 detecting detect VERB fcis-28493 188 19 grasp grasp NOUN fcis-28493 188 20 configurations configuration NOUN fcis-28493 188 21 from from ADP fcis-28493 188 22 point point NOUN fcis-28493 188 23 sets[c]//2019 sets[c]//2019 PUNCT fcis-28493 188 24 international international ADJ fcis-28493 188 25 conference conference NOUN fcis-28493 188 26 on on ADP fcis-28493 188 27 robotics robotic NOUN fcis-28493 188 28 and and CCONJ fcis-28493 188 29 automation automation NOUN fcis-28493 188 30 ( ( PUNCT fcis-28493 188 31 icra icra PROPN fcis-28493 188 32 ) ) PUNCT fcis-28493 188 33 . . PUNCT fcis-28493 189 1 ieee ieee PROPN fcis-28493 189 2 , , PUNCT fcis-28493 189 3 2019 2019 NUM fcis-28493 189 4 : : PUNCT fcis-28493 189 5 3629 3629 NUM fcis-28493 189 6 - - SYM fcis-28493 189 7 3635 3635 NUM fcis-28493 189 8 . . PUNCT fcis-28493 190 1 [ [ X fcis-28493 190 2 4 4 NUM fcis-28493 190 3 ] ] X fcis-28493 190 4 mousavian mousavian NOUN fcis-28493 190 5 a a PROPN fcis-28493 190 6 , , PUNCT fcis-28493 190 7 eppner eppner NOUN fcis-28493 190 8 c c NOUN fcis-28493 190 9 , , PUNCT fcis-28493 190 10 fox fox PROPN fcis-28493 190 11 d. d. PROPN fcis-28493 190 12 6 6 NUM fcis-28493 190 13 - - PUNCT fcis-28493 190 14 dof dof NOUN fcis-28493 190 15 graspnet graspnet NOUN fcis-28493 190 16 : : PUNCT fcis-28493 190 17 variational variational ADJ fcis-28493 190 18 grasp grasp NOUN fcis-28493 190 19 generation generation NOUN fcis-28493 190 20 for for ADP fcis-28493 190 21 object object NOUN fcis-28493 190 22 manipulation[c]//proceedings manipulation[c]//proceeding NOUN fcis-28493 190 23 of of ADP fcis-28493 190 24 the the DET fcis-28493 190 25 ieee/ ieee/ NUM fcis-28493 190 26 cvf cvf NOUN fcis-28493 190 27 international international ADJ fcis-28493 190 28 conference conference NOUN fcis-28493 190 29 on on ADP fcis-28493 190 30 computer computer NOUN fcis-28493 190 31 vision vision NOUN fcis-28493 190 32 . . PUNCT fcis-28493 191 1 2019 2019 NUM fcis-28493 191 2 : : PUNCT fcis-28493 191 3 2901 2901 NUM fcis-28493 191 4 - - SYM fcis-28493 191 5 2910 2910 NUM fcis-28493 191 6 . . PUNCT fcis-28493 192 1 [ [ X fcis-28493 192 2 5 5 X fcis-28493 192 3 ] ] X fcis-28493 192 4 qin qin PROPN fcis-28493 192 5 y y PROPN fcis-28493 192 6 , , PUNCT fcis-28493 192 7 chen chen PROPN fcis-28493 192 8 r r PROPN fcis-28493 192 9 , , PUNCT fcis-28493 192 10 zhu zhu PROPN fcis-28493 192 11 h h NOUN fcis-28493 192 12 , , PUNCT fcis-28493 193 1 et et PROPN fcis-28493 193 2 al al PROPN fcis-28493 193 3 . . PROPN fcis-28493 193 4 s4 s4 PROPN fcis-28493 193 5 g g NOUN fcis-28493 193 6 : : PUNCT fcis-28493 193 7 amodal amodal ADJ fcis-28493 193 8 single single ADJ fcis-28493 193 9 - - PUNCT fcis-28493 193 10 view view NOUN fcis-28493 193 11 singleshot singleshot NOUN fcis-28493 193 12 se se PROPN fcis-28493 193 13 ( ( PUNCT fcis-28493 193 14 3 3 X fcis-28493 193 15 ) ) PUNCT fcis-28493 193 16 grasp grasp NOUN fcis-28493 193 17 detection detection NOUN fcis-28493 193 18 in in ADP fcis-28493 193 19 cluttered cluttered ADJ fcis-28493 193 20 scenes[c]//conference scenes[c]//conference NOUN fcis-28493 193 21 on on ADP fcis-28493 193 22 robot robot NOUN fcis-28493 193 23 learning learning NOUN fcis-28493 193 24 . . PUNCT fcis-28493 194 1 pmlr pmlr NOUN fcis-28493 194 2 , , PUNCT fcis-28493 194 3 2020 2020 NUM fcis-28493 194 4 : : PUNCT fcis-28493 194 5 53 53 NUM fcis-28493 194 6 - - SYM fcis-28493 194 7 65 65 NUM fcis-28493 194 8 . . PUNCT fcis-28493 195 1 [ [ X fcis-28493 195 2 6 6 NUM fcis-28493 195 3 ] ] PUNCT fcis-28493 195 4 qi qi PROPN fcis-28493 195 5 c c PROPN fcis-28493 195 6 r r PROPN fcis-28493 195 7 , , PUNCT fcis-28493 195 8 yi yi NOUN fcis-28493 195 9 l l NOUN fcis-28493 195 10 , , PUNCT fcis-28493 195 11 su su PROPN fcis-28493 195 12 h h PROPN fcis-28493 195 13 , , PUNCT fcis-28493 195 14 et et PROPN fcis-28493 195 15 al al PROPN fcis-28493 195 16 . . PROPN fcis-28493 195 17 pointnet++ pointnet++ PROPN fcis-28493 195 18 : : PUNCT fcis-28493 195 19 deep deep ADJ fcis-28493 195 20 hierarchical hierarchical ADJ fcis-28493 195 21 feature feature NOUN fcis-28493 195 22 learning learn VERB fcis-28493 195 23 on on ADP fcis-28493 195 24 point point NOUN fcis-28493 195 25 sets set NOUN fcis-28493 195 26 in in ADP fcis-28493 195 27 a a DET fcis-28493 195 28 metric metric ADJ fcis-28493 195 29 space[j space[j NOUN fcis-28493 195 30 ] ] PUNCT fcis-28493 195 31 . . PUNCT fcis-28493 196 1 advances advance NOUN fcis-28493 196 2 in in ADP fcis-28493 196 3 neural neural ADJ fcis-28493 196 4 information information NOUN fcis-28493 196 5 processing processing NOUN fcis-28493 196 6 systems system NOUN fcis-28493 196 7 , , PUNCT fcis-28493 196 8 2017 2017 NUM fcis-28493 196 9 , , PUNCT fcis-28493 196 10 30 30 NUM fcis-28493 196 11 . . PUNCT fcis-28493 197 1 [ [ X fcis-28493 197 2 7 7 NUM fcis-28493 197 3 ] ] PUNCT fcis-28493 197 4 xu xu PROPN fcis-28493 197 5 j j PROPN fcis-28493 197 6 , , PUNCT fcis-28493 197 7 pan pan PROPN fcis-28493 197 8 y y PROPN fcis-28493 197 9 , , PUNCT fcis-28493 197 10 pan pan NOUN fcis-28493 197 11 x x PROPN fcis-28493 197 12 , , PUNCT fcis-28493 197 13 et et PROPN fcis-28493 197 14 al al PROPN fcis-28493 197 15 . . PUNCT fcis-28493 198 1 regnet regnet PROPN fcis-28493 198 2 : : PUNCT fcis-28493 198 3 self self NOUN fcis-28493 198 4 - - PUNCT fcis-28493 198 5 regulated regulate VERB fcis-28493 198 6 network network NOUN fcis-28493 198 7 for for ADP fcis-28493 198 8 image image NOUN fcis-28493 198 9 classification[j classification[j NOUN fcis-28493 198 10 ] ] PUNCT fcis-28493 198 11 . . PUNCT fcis-28493 199 1 ieee ieee NOUN fcis-28493 199 2 transactions transaction NOUN fcis-28493 199 3 on on ADP fcis-28493 199 4 neural neural ADJ fcis-28493 199 5 networks network NOUN fcis-28493 199 6 and and CCONJ fcis-28493 199 7 learning learning NOUN fcis-28493 199 8 systems system NOUN fcis-28493 199 9 , , PUNCT fcis-28493 199 10 2022 2022 NUM fcis-28493 199 11 , , PUNCT fcis-28493 199 12 34(11 34(11 NUM fcis-28493 199 13 ): ): PUNCT fcis-28493 199 14 9562 9562 NUM fcis-28493 199 15 - - SYM fcis-28493 199 16 9567 9567 NUM fcis-28493 199 17 . . PUNCT fcis-28493 200 1 [ [ X fcis-28493 200 2 8 8 X fcis-28493 200 3 ] ] X fcis-28493 200 4 huang huang PROPN fcis-28493 200 5 h h PROPN fcis-28493 200 6 , , PUNCT fcis-28493 200 7 wang wang PROPN fcis-28493 200 8 d d PROPN fcis-28493 200 9 , , PUNCT fcis-28493 200 10 zhu zhu PROPN fcis-28493 200 11 x x X fcis-28493 200 12 , , PUNCT fcis-28493 200 13 et et PROPN fcis-28493 200 14 al al PROPN fcis-28493 200 15 . . PROPN fcis-28493 200 16 edge edge PROPN fcis-28493 200 17 grasp grasp NOUN fcis-28493 200 18 network network NOUN fcis-28493 200 19 : : PUNCT fcis-28493 200 20 a a DET fcis-28493 200 21 graphbased graphbased ADJ fcis-28493 200 22 se se X fcis-28493 200 23 ( ( PUNCT fcis-28493 200 24 3)-invariant 3)-invariant ADJ fcis-28493 200 25 approach approach NOUN fcis-28493 200 26 to to PART fcis-28493 200 27 grasp grasp VERB fcis-28493 200 28 detection[c]//2023 detection[c]//2023 PROPN fcis-28493 200 29 ieee ieee NOUN fcis-28493 200 30 international international ADJ fcis-28493 200 31 conference conference NOUN fcis-28493 200 32 on on ADP fcis-28493 200 33 robotics robotic NOUN fcis-28493 200 34 and and CCONJ fcis-28493 200 35 automation automation NOUN fcis-28493 200 36 ( ( PUNCT fcis-28493 200 37 icra icra PROPN fcis-28493 200 38 ) ) PUNCT fcis-28493 200 39 . . PUNCT fcis-28493 201 1 ieee ieee NOUN fcis-28493 201 2 , , PUNCT fcis-28493 201 3 2023 2023 NUM fcis-28493 201 4 : : PUNCT fcis-28493 201 5 3882 3882 NUM fcis-28493 201 6 - - SYM fcis-28493 201 7 3888 3888 NUM fcis-28493 201 8 . . PUNCT fcis-28493 202 1 [ [ X fcis-28493 202 2 9 9 NUM fcis-28493 202 3 ] ] X fcis-28493 202 4 hamilton hamilton PROPN fcis-28493 202 5 w w PROPN fcis-28493 202 6 , , PUNCT fcis-28493 202 7 ying ying PROPN fcis-28493 202 8 z z PROPN fcis-28493 202 9 , , PUNCT fcis-28493 202 10 leskovec leskovec ADP fcis-28493 202 11 j. j. PROPN fcis-28493 202 12 inductive inductive PROPN fcis-28493 202 13 representation representation NOUN fcis-28493 202 14 learning learn VERB fcis-28493 202 15 on on ADP fcis-28493 202 16 large large ADJ fcis-28493 202 17 graphs[j graphs[j PROPN fcis-28493 202 18 ] ] PUNCT fcis-28493 202 19 . . PUNCT fcis-28493 203 1 advances advance NOUN fcis-28493 203 2 in in ADP fcis-28493 203 3 neural neural ADJ fcis-28493 203 4 information information NOUN fcis-28493 203 5 processing processing NOUN fcis-28493 203 6 systems system NOUN fcis-28493 203 7 , , PUNCT fcis-28493 203 8 2017 2017 NUM fcis-28493 203 9 , , PUNCT fcis-28493 203 10 30 30 NUM fcis-28493 203 11 . . PUNCT fcis-28493 204 1 [ [ X fcis-28493 204 2 10 10 NUM fcis-28493 204 3 ] ] X fcis-28493 204 4 xiang xiang PROPN fcis-28493 204 5 y y PROPN fcis-28493 204 6 , , PUNCT fcis-28493 204 7 schmidt schmidt PROPN fcis-28493 204 8 t t PROPN fcis-28493 204 9 , , PUNCT fcis-28493 204 10 narayanan narayanan PROPN fcis-28493 204 11 v v PROPN fcis-28493 204 12 , , PUNCT fcis-28493 204 13 et et PROPN fcis-28493 204 14 al al PROPN fcis-28493 204 15 . . PROPN fcis-28493 205 1 posecnn posecnn PROPN fcis-28493 205 2 : : PUNCT fcis-28493 205 3 a a DET fcis-28493 205 4 convolutional convolutional ADJ fcis-28493 205 5 neural neural ADJ fcis-28493 205 6 network network NOUN fcis-28493 205 7 for for ADP fcis-28493 205 8 6d 6d NUM fcis-28493 205 9 object object NOUN fcis-28493 205 10 pose pose NOUN fcis-28493 205 11 estimation estimation NOUN fcis-28493 205 12 in in ADP fcis-28493 205 13 cluttered cluttered ADJ fcis-28493 205 14 scenes[j scenes[j NOUN fcis-28493 205 15 ] ] PUNCT fcis-28493 205 16 . . PUNCT fcis-28493 206 1 arxiv arxiv PROPN fcis-28493 206 2 preprint preprint PROPN fcis-28493 206 3 arxiv:1711.00199 arxiv:1711.00199 NOUN fcis-28493 206 4 , , PUNCT fcis-28493 206 5 2017 2017 NUM fcis-28493 206 6 . . PUNCT fcis-28493 207 1 [ [ X fcis-28493 207 2 11 11 NUM fcis-28493 207 3 ] ] X fcis-28493 207 4 singh singh PROPN fcis-28493 207 5 a a PROPN fcis-28493 207 6 , , PUNCT fcis-28493 207 7 sha sha PROPN fcis-28493 207 8 j j PROPN fcis-28493 207 9 , , PUNCT fcis-28493 207 10 narayan narayan PROPN fcis-28493 207 11 k k PROPN fcis-28493 207 12 s s PROPN fcis-28493 207 13 , , PUNCT fcis-28493 207 14 et et PROPN fcis-28493 207 15 al al PROPN fcis-28493 207 16 . . PROPN fcis-28493 207 17 bigbird:(big bigbird:(big PROPN fcis-28493 207 18 ) ) PUNCT fcis-28493 207 19 berkeley berkeley PROPN fcis-28493 207 20 instance instance PROPN fcis-28493 207 21 recognition recognition PROPN fcis-28493 207 22 dataset[c]//2014 dataset[c]//2014 PROPN fcis-28493 207 23 ieee ieee PROPN fcis-28493 207 24 international international PROPN fcis-28493 207 25 conference conference NOUN fcis-28493 207 26 on on ADP fcis-28493 207 27 robotics robotic NOUN fcis-28493 207 28 and and CCONJ fcis-28493 207 29 automation automation NOUN fcis-28493 207 30 . . PUNCT fcis-28493 208 1 2014 2014 NUM fcis-28493 208 2 : : PUNCT fcis-28493 208 3 509 509 NUM fcis-28493 208 4 - - SYM fcis-28493 208 5 516 516 NUM fcis-28493 208 6 . . PUNCT fcis-28493 209 1 [ [ X fcis-28493 209 2 12 12 NUM fcis-28493 209 3 ] ] PUNCT fcis-28493 209 4 kasper kasper NOUN fcis-28493 209 5 a a PROPN fcis-28493 209 6 , , PUNCT fcis-28493 209 7 xue xue PROPN fcis-28493 209 8 z z PROPN fcis-28493 209 9 , , PUNCT fcis-28493 209 10 dillmann dillmann PROPN fcis-28493 209 11 r. r. PROPN fcis-28493 209 12 the the DET fcis-28493 209 13 kit kit NOUN fcis-28493 209 14 object object NOUN fcis-28493 209 15 models model NOUN fcis-28493 209 16 database database VERB fcis-28493 209 17 : : PUNCT fcis-28493 209 18 an an DET fcis-28493 209 19 object object NOUN fcis-28493 209 20 model model NOUN fcis-28493 209 21 database database NOUN fcis-28493 209 22 for for ADP fcis-28493 209 23 object object NOUN fcis-28493 209 24 recognition recognition NOUN fcis-28493 209 25 , , PUNCT fcis-28493 209 26 localization localization NOUN fcis-28493 209 27 and and CCONJ fcis-28493 209 28 manipulation manipulation NOUN fcis-28493 209 29 in in ADP fcis-28493 209 30 service service NOUN fcis-28493 209 31 robotics[j robotics[j PROPN fcis-28493 209 32 ] ] PUNCT fcis-28493 209 33 . . PUNCT fcis-28493 210 1 the the DET fcis-28493 210 2 international international ADJ fcis-28493 210 3 journal journal NOUN fcis-28493 210 4 of of ADP fcis-28493 210 5 robotics robotic NOUN fcis-28493 210 6 research research NOUN fcis-28493 210 7 , , PUNCT fcis-28493 210 8 2012 2012 NUM fcis-28493 210 9 , , PUNCT fcis-28493 210 10 31(8 31(8 NUM fcis-28493 210 11 ): ): PUNCT fcis-28493 210 12 927 927 NUM fcis-28493 210 13 - - SYM fcis-28493 210 14 934 934 NUM fcis-28493 210 15 . . PUNCT fcis-28493 211 1 [ [ X fcis-28493 211 2 13 13 NUM fcis-28493 211 3 ] ] X fcis-28493 211 4 breyer breyer NOUN fcis-28493 211 5 m m PROPN fcis-28493 211 6 , , PUNCT fcis-28493 211 7 chung chung PROPN fcis-28493 211 8 j j PROPN fcis-28493 211 9 j j PROPN fcis-28493 211 10 , , PUNCT fcis-28493 211 11 ott ott PROPN fcis-28493 211 12 l l PROPN fcis-28493 211 13 , , PUNCT fcis-28493 211 14 et et PROPN fcis-28493 212 1 al al PROPN fcis-28493 212 2 . . PROPN fcis-28493 212 3 volumetric volumetric NOUN fcis-28493 212 4 grasping grasp VERB fcis-28493 212 5 network network NOUN fcis-28493 212 6 : : PUNCT fcis-28493 212 7 real real ADJ fcis-28493 212 8 - - PUNCT fcis-28493 212 9 time time NOUN fcis-28493 212 10 6 6 NUM fcis-28493 212 11 dof dof NOUN fcis-28493 212 12 grasp grasp NOUN fcis-28493 212 13 detection detection NOUN fcis-28493 212 14 in in ADP fcis-28493 212 15 clutter[c]//conference clutter[c]//conference PROPN fcis-28493 212 16 on on ADP fcis-28493 212 17 robot robot NOUN fcis-28493 212 18 learning learning NOUN fcis-28493 212 19 . . PUNCT fcis-28493 213 1 pmlr pmlr NOUN fcis-28493 213 2 , , PUNCT fcis-28493 213 3 2021 2021 NUM fcis-28493 213 4 : : PUNCT fcis-28493 213 5 1602 1602 NUM fcis-28493 213 6 - - SYM fcis-28493 213 7 1611 1611 NUM fcis-28493 213 8 . . PUNCT