ビンピッキングのためのRGB-Dカメラを用いた三次元位置姿勢推定，および把持可能性を考慮したスコアリング手法

全文

(1)Vol.2014-CG-157 No.2 Vol.2014-CVIM-194 No.2 2014/11/20. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report. ϏϯϐοΩϯάͷͨΊͷ RGB-D ΧϝϥΛ༻͍ͨࡾ࣍‫ݩ‬Ґஔ ࢟੎ਪఆɼ͓Αͼ೺࣋ՄೳੑΛߟྀͨ͠είΞϦϯάख๏ ੢ ୎࿠1,a). ٢‫ོ ݟ‬1. ߴ੉ ཽҰ1. ‫ݚ ాݪ‬հ1. Ӭా ࿨೭1. ৽ྑ‫ و‬ཅฏ2. ՏҪ ྑߒ1. ֓ཁɿϗϏʔ༻ͱͯ͠ී‫͍ͯ͠ٴ‬Δ҆Ձͳ RGB-D ΧϝϥΑΓಘΒΕΔ‫཭ڑ‬ը૾Λର৅ͱͯ͠ɼେྔͷࡾ ࣍‫ݩ‬ಛ௃఺Λ༻͍Δ͜ͱͰόϥੵ෺ମͷ‫ݸ‬ʑͷҐஔ࢟੎Λ҆ఆͯ͠‫ݕ‬ग़͢Δख๏ΛఏҊ͢ΔɽϥϯμϜα ϯϓϦϯάʹՃ͑ CAD Ϟσϧ͔Β͋Β͔͡Ίྨࣅ࢟੎Λ‫Ͱͱ͓ͯ͘͜͠ࢉܭ‬ɼେ‫ہ‬త࠷దղΛਪఆՄೳ ͱͳͬͨɽ·ͨɼCAD Ϟσϧʹ೺࣋఺ͷ৘ใΛ෇Ճ͢Δ͜ͱʹΑΓɼର৅෺ۙลͷϚχϐϡϨʔλͷ‫ׯ‬ব ۭؒΛ‫͢ࢉܭ‬Δ͜ͱͰෳ਺ͷର৅෺ͷத͔Β࣮ࡍʹ೺࣋Մೳͳର৅෺ͷҐஔ࢟੎ɼ೺࣋఺ͷҐஔɼΤϯυ ΤϑΣΫλͷ࢟੎Λܾఆ͢Δ͜ͱ͕ՄೳͱͳΓɼෳࡶͳ‫ڠ‬ௐ੍‫ޚ‬Λߦ͏͜ͱͳ͠ʹ࣮༻తͳϏϯϐοΩϯ άγεςϜ͕ߏஙͰ͖Δ͜ͱΛࣔͨ͠ɽ. 1. ͸͡Ίʹ ੡଄‫ݱ‬৔ʹ͓͚ΔϏϯϐοΩϯάͰ͸ɼର৅෺ͷҐஔ࢟ ੎ͷ‫ݕ‬ग़Λࢹ֮෦ʹཁ‫͞ٻ‬ΕΔ͜ͱ͕ଟ͍ɽ͜ΕʹԠ͑Δ ͨΊɼ‫ʹͰ·ࡏݱ‬ɼର৅෺ͷ‫ڥ‬քઢΛಛ௃ྔͱͯ͠༻͍Δ ख๏ [1][2] ΍ɼ఺‫܈‬ͷ‫ॴہ‬ಛ௃ྔΛ༻͍Δख๏ [3][4][5][6]. ͯόϥੵΈ͞Εͨ෺ମͷ‫ݸ‬ʑͷҐஔ࢟੎Λ‫ݕ‬ग़Ͱ͖ΔΞϧ ΰϦζϜͷ։ൃ͠ɼ‫ݕ‬ग़ͨ͠ର৅෺ͷҐஔ࢟੎͕Ϛχϐϡ ϨʔλͰ҆ఆͯ͠೺࣋Մೳͳ͜ͱΛߟྀͨ͠είΞϦϯά ख๏ΛఏҊ͢Δɽ. 2. λεΫͱγεςϜߏ੒. ͳͲ͕ఏএ͞ΕɼҰఆͷ੒ՌΛ্͍͛ͯΔɽ͔͜͠͠ΕΒ. ຊ‫Ͱڀݚ‬͸੡଄‫ݱ‬৔͔Βͷཁ੥ʹ‫͖ͮج‬ɼόϥੵΈͷঢ়. ͷख๏ͷଟ͘͸ɼγʔϯͷࡾ࣍‫ܗݩ‬ঢ়͕͋Δఔ౓҆ఆͯ͠. ଶͷ 5 छྨͷ‫ܗ‬ঢ়ͷ෦඼ (ਤ 1-a.) Λ࣏۩্ (ਤ 1-b.) ʹ૊. ಘΒΕΔ͜ͱΛલఏͱ͓ͯ͠ΓɼෳࡶͰߴՁͳࡱ૾૷ஔɼ. Έ্͛ɼਤ 1-c. ͷঢ়ଶʹ͢Δ͜ͱΛλεΫͱͯ͠ઃఆͨ͠ɽ. ઃඋͷಋೖ΍‫ڥ؀‬ͷ੔උ͕ඞཁͱͳͬͯɼಋೖɼӡ༻ʹ͔ ͔Δίετ͕େ͖͘ͳΔ໰୊͕͋Δɽ. 3 छྨͷύΠϓ͓Αͼ 2 छྨͷϓϨʔτ͔ΒͳΔۚଐ੡ ͷ֤ର৅෦඼͸ࠞࡏ͢Δ͜ͱͳ͘ผʑͷശʹόϥੵΈͰ. ·ͨɼ࣮ࡍͷγεςϜʹ͓͍ͯ͸ɼ෺ମͷҐஔ࢟੎ʹՃ. ೖΕΒΕ͓ͯΓɼ͜ΕΛ্ํ͔Β RGB-D Χϝϥ (ਤ 1-1.). ͑ɼͦΕ͕҆ఆͯ͠೺࣋ՄೳͰ͋Δ͔Ͳ͏͔ͷ‫͕ূݕ‬ඞཁ. ʹΑͬͯࡱ૾ͯ͠Ґஔ࢟੎ͷਪఆΛߦ͍ɼͦͷ݁Ռʹ‫ج‬. ͱ͞ΕΔɽ͜Εʹؔͯ͠͸‫ʹͰ·ࡏݱ‬ɼϚχϐϡϨʔγϣ. ͍ͮͯ૒࿹ϩϘοτͷӈ࿹ʹऔΓ෇͚ΒΕͨ‫ٵ‬ணύου. ϯͷ໰୊ͱͯ͠ɼ೺࣋Մೳ఺ͷ୳ࡧʹΑΓಈ࡞‫ܭ‬ըΛߦ. (ਤ 1-2.) ʹΑΓ෦඼ΛऔΓग़͠ɼࠨखͷ૊Έ෇͚༻ࡾࢦϋ. ͏ [7]ɼର৅෺Λԁ౵Ͱۙࣅ͠‫ׯ‬বνΣοΫΛߦ͏ [8] ͳͲ. ϯυ (ਤ 1-3.) ʹ࣋ͪସ্͑ͨͰ࣏۩΁ͷ૊Έ෇͚Λߦ͏ɽ. ͷΞϓϩʔν͕ͳ͞Ε͖͍ͯͯΔ͕ɼ‫ʹ࣮ݱ‬͸࣮૷ͷࠔ೉ ͞ͳͲ͔ΒώϡʔϦεςΟοΫͳख๏Ͱղܾ͞ΕΔ͜ͱ΋ ଟ͍ [9][10]ɽ͔͠͠ର৅෺ͷҐஔ࢟੎‫ݕ‬ग़ͱಉ࣌ʹपғʹ ଘࡏ͢Δো֐෺ͷҐஔ৘ใ΋औಘՄೳͰ͋Δ͜ͱ͔Βɼ೺ ࣋Մೳੑʹ͍ͭͯ΋ϏδϣϯଆͰॲཧ͢Δ͜ͱͰ͜ͷ໰୊ Λޮ཰తʹղܾͰ͖Δ΋ͷͱ‫ظ‬଴Ͱ͖Δɽ. ຊใͰ͸͜ΕΒҰ࿈ͷλεΫͷ͏ͪɼಛʹόϥੵΈঢ়ଶ ͷ෦඼ͷऔΓग़͠ʹؔ͢Δ޻ఔʹ͍ͭͯड़΂Δɽ. 3. όϥੵΈ෺ମͷҐஔ࢟੎ਪఆ όϥੵΈঢ়ଶͷର৅෺ͷࡾ࣍‫ܗݩ‬ঢ়͸ࢢൢͷ҆Ձͳ RGB-. D ΧϝϥΛ༻͍ͯ‫ܭ‬ଌ͢Δɽ͔͜͠͠ͷ‫཭ڑ‬ը૾͸ϊΠζ. ຊใͰ͸ɼϗϏʔ༻ͱͯ͠ී‫͍ͯ͠ٴ‬Δ҆Ձͳ RGB-D. ΍ܽམΛଟ͘‫ؚ‬Έɼैདྷͷख๏Ͱ͸ϚχϐϡϨʔλʹΑΔ. ΧϝϥΛ༻͍ͯऔಘͨ͠ϊΠζͷଟ͍‫཭ڑ‬ը૾͔Β҆ఆ͠. ೺࣋Λߦ͏ͨΊͷਫ਼౓ͰҐஔ࢟੎ͷਪఆΛߦ͏͜ͱ͕ࠔ೉. 1 2 a). Ͱ͋ͬͨɽຊ‫Ͱڀݚ‬͸ɼ‫཭ڑ‬ը૾ʹରͯ͠େ͖ͳ΢Οϯυ ಠཱߦ੓๏ਓ࢈‫ٕۀ‬ज़૯߹‫ॴڀݚ‬ AIST, Tsukuba, Ibaraki 305–8586 Japan ౦‫ࢢ౎ژ‬େֶେֶӃ Tokyo City University, Setagaya, Tokyo 158–0087 Japan [email protected]. ⓒ 2014 Information Processing Society of Japan. ΢Λ࢖༻ͯ͠ೋ࣍‫ۂ‬໘Λ౰ͯ͸ΊΔ͜ͱͰϊΠζ΍ܽམͷ վળΛਤΔͱ‫ʹڞ‬ɼࡾ࣍‫ݩ‬ಛ௃ྔͱͯ͠ɼγʔϯ͓ΑͼϞ σϧΛߏ੒͢Δ֤఺ͷࡾ࣍‫࠲ݩ‬ඪ஋ͱɼ‫؍‬ଌํ޲ʹґଘ͠. 1.

(2) Vol.2014-CG-157 No.2 Vol.2014-CVIM-194 No.2 2014/11/20. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report. xi = Ai s2 + Bi t2 + Ci st + Di s + Ei t + Fi. 2. Right Hand 1. RGB-D camera. (2). ͜͜Ͱɼxi (i : {0, 1, 2}) ͸γʔϯ֤఺ͷࡾ࣍‫࠲ݩ‬ඪ஋ɼs, t ͸ೋ࣍‫ݩ‬ը૾্ͷ col, row ࠲ඪ஋Λද͢ɽ. 3. Left Hand. ͜ͷॲཧ͸ɼಛ௃ྔͷ‫ͱࢉܭ‬ಉ࣌ʹγʔϯதͷϊΠζ΍ ܽଛΛ௿‫͢ݮ‬ΔޮՌ΋࣋ͭɽͨͩ͠΢Οϯυ΢ॲཧʹΑΓ ෺ମͷ‫ڥ‬ք෇ۙͷ‫ܗ‬ঢ়͕େ͖͘มԽͯ͠͠·͏͜ͱΛආ͚ ΔͨΊɼ‫ۂ‬໘͋ͯ͸Ί͸ɼ͋Β͔͡Ίೋ࣍‫ݩ‬ը૾্Ͱ‫཭ڑ‬ ΤοδΛ༻͍ͯྖҬ෼ׂॲཧΛߦ͍ɼ஫໨ըૉͱಉҰྖҬ ʹଐ͢Δ΢Οϯυ΢಺ͷ఺͚ͩΛ࢖ͬͯߦ͏͜ͱͱͨ͠ɽ ͳ͓΢Οϯυ΢αΠζ͸ɼϞσϧͷ୅දతͳ‫཰ۂ‬൒‫ͱܘ‬ೖ ྗσόΠεͷಛੑ͔Βܾఆ͢Δ͜ͱͱͨ͠ɽ. 3.2.2 γʔϯͷྖҬ෼ׂ Ϟσϧর߹ͷର৅ͱͳΔγʔϯதͷ఺ͷ਺Λ੍‫͢ݶ‬Δͨ a. Target Objects a. Ίʹɼઌड़ͷ‫཭ڑ‬ΤοδʹՃ͑ɼओํ޲ϕΫτϧΤοδ͓. b Jig b.. Αͼ‫ۂ‬໘ͷ෼ྨΛ༻͍ͯγʔϯͷྖҬ෼ׂΛߦ͏ɽͨͩ͠ ‫ۂ‬໘ͷ෼ྨ͸ɼҰൠతͳΨ΢ε‫཰ۂ‬ɼฏ‫཰ۂۉ‬Λ༻͍Δํ ๏Ͱ͸ແ͘ɼShape Index[12] ͱ‫ݺ‬͹ΕΔํ๏ʹ‫͖ͮج‬ɼ࠷ େ‫ ཰ۂ‬Cmax ɼ࠷খ‫ ཰ۂ‬Cmin Λࣜ (3) ʹΑΓ‫࠲ۃ‬ඪม‫͠׵‬ ͯಘΒΕΔ Cr ͓Αͼ Cθ ͔Βɼತ: −3/4π ≤ Cθ < −π/4ɼ Ҍ‫ܕ‬: −π/2 ≤ Cθ ≤ 0ɼԜ: −π/4 < Cθ ≤ π/4ɼฏ໘: c. Assembled object ਤ 1. 0 ≤ Cr ≤ T hF ͱͨ͠ɽ͜͜Ͱ T hF ͸ೖྗσόΠεͷಛ ੑʹ߹Θͤͨᮢ஋Ͱ͋Γɼ֓Ͷ 0.002 ≤ T hF ≤ 0.01 ఔ౓. γεςϜશମਤ. Fig. 1 Systems overview. ͜ͱͰɼϊΠζʹରͯ͠‫ͳͱڧؤ‬ΔϞσϧর߹ΛՄೳͱ͢. ͷ஋ΛऔΔɽ ⎧ C ⎪ ⎨Cθ = tan−1 min Cmax ⎪ ⎩ 2 2 Cr = Cmax + Cmin. Δख๏Λ։ൃͨ͠ [11]ɽ. ͜ΕʹΑΓɼγʔϯதͷ 1 ఺͸ɼ࠷େ 3 ͭͷྖҬʹಉ࣌ʹ. ͳ͍ෆมಛ௃ྔͱͯ͠‫͔͘ݹ‬Β஌ΒΕΔɼओํ޲ϕΫτϧ ͓Αͼ࠷େ‫཰ۂ‬ɼ࠷খ‫཰ۂ‬ͷ 14 ࣍‫ݩ‬ͷϕΫτϧΛ༻͍Δ. (3). ॴଐ͢Δ͜ͱʹͳΔɽ. 3.1 ೝࣝϞσϧͷ࡞੒ STL ʹ୅ද͞ΕΔ CAD Ϟσϧ͸Ұൠతʹ఺ͷີ౓͕Ұ ༷Ͱ͸ແ͍ɽຊใͰ͸ɼCAD ϞσϧΛҰఆີ౓ͷ఺‫܈‬σʔ. 3.3 Ϟσϧর߹ ࡾ࣍‫࠲ݩ‬ඪͱओํ޲ϕΫτϧ͕‫ط‬஌ͷ৔߹ɼϞσϧ্ͷ. λͱͯ͠ల։͠ɼϞσϧͷෳࡶ͞ʹԠͯ͡ఆΊΔ൒‫ࢠྔ( ܘ‬. Ұ఺ (Ϟσϧ఺) ͱγʔϯதͷҰ఺ (γʔϯ఺) ͷ૊Έ߹Θ. Խϐον) Λ࣋ͭ΢Οϯυ΢಺ʹ͋Δ఺Λର৅ͱͯ͠ɼओ. ͤʢϖΞʣΛܾఆ͢Δ͜ͱͰɼϞσϧͷҐஔ࢟੎Λ࠷େ/. ੒෼෼ੳΛߦ͏͜ͱͰୈ 1 ओ੒෼ uɼୈ 2 ओ੒෼ v ɼ͓Α. ࠷খ‫޲ํ཰ۂ‬ϕΫτϧͷූ߸ํ޲Λআ͍ͯҰҙʹܾఆͰ͖. ͼ uɼv ͷ֎ੵ w Λ‫ٻ‬Ίͯࣜ (1) ͷ‫ۂ‬໘΁ͷ͋ͯ͸ΊΛߦ. ΔɽఏҊख๏Ͱ͸͜ͷ͜ͱΛར༻͠ɼγʔϯͷྖҬຖʹ࠷. ͍ɼओํ޲ϕΫτϧ͓Αͼ࠷େ/࠷খ‫཰ۂ‬ͷ‫ࢉܭ‬Λߦͬͨɽ. w = Au2 + Bv 2 + Cuv + Du + Ev + F. (1). దͳϖΞ (γʔυϖΞ) Λ RANSAC[13] ʹΑΓ‫ٻ‬ΊΔ͜ͱ ͰϞσϧর߹Λߦ͏ɽҐஔ࢟੎ͷධՁ͸ɼγʔυϖΞʹΑ Γܾఆ͞ΕͨҐஔ࢟੎ΛͱΔϞσϧͷશ఺Λγʔϯʹ౤Ө. ΢Οϯυ΢಺ʹଘࡏ͢Δ఺͕Ұఆ਺ΑΓগͳ͍ɼ·ͨ͸͋. ͠ɼa) ๏ઢํ޲͕ࢹઢํ޲ʹରͯ͠ 90 ౓Ҏ಺ɼb) ࡾ࣍‫ݩ‬. ͯ͸Ί‫͕ࠩޡ‬Ұఆ஋ΑΓେ͖͍৔߹ʹ͸ɼܽଛ஋ͱͯ͠র. ࠲ඪ஋͕౤Өઌγʔϯ఺ͱҰఆ‫཭ڑ‬Ҏ಺ɼc) ओํ޲ϕΫτ. ߹ʹ༻͍ͳ͍Α͏ʹͨ͠ɽ. ϧ͓Αͼ‫཰ۂ‬ͷ౤Өઌγʔϯ఺ͱͷίαΠϯ‫͕཭ڑ‬Ұఆ஋ Ҏ্ɼͷࡾͭͷ৚݅Λ͢΂ͯຬͨ͢఺Λ਺্͑͛Δ͜ͱͰ. 3.2 γʔϯͷ࡞੒. ߦͬͨɽͨͩ͠γʔυϖΞΛߏ੒͢ΔϞσϧ఺͸γʔϯ఺. 3.2.1 ಛ௃ྔͷ‫ࢉܭ‬. ͱಉ͡‫ۂ‬໘ͷ‫ܕ‬Λ࣋ͭ΋ͷ͔ΒબͿΑ͏ʹͨ͠ɽ·ͨධՁ. γʔϯ֤఺ͷओํ޲ϕΫτϧɼ͓Αͼ࠷େ/࠷খ‫཰ۂ‬͸ɼ. ஋͕Ұఆ஋Ҏ্ͷγʔυʹ͍ͭͯɼ‫ࡏݱ‬ͷγʔϯ఺ͷपล. ‫཭ڑ‬ը૾্Ͱ΢Οϯυ΢ॲཧΛ༻͍ɼࣜ (2) Ͱද͞ΕΔύ. ͰධՁ஋͕ΑΓߴ͘ͳΔ఺Λ࠶‫ؼ‬తʹ୳ࡧ͢Δ͜ͱͰۙ๣. ϥϝτϦοΫೋ࣍‫ۂ‬໘Λ͋ͯ͸ΊΔ͜ͱͰ‫ٻ‬Ίͨɽ. ͷ఺ͷத͔Β࠷దͳ࢟੎Λ‫ٻ‬ΊΔΑ͏ʹͨ͠ɽ. ⓒ 2014 Information Processing Society of Japan. 2.

(3) Vol.2014-CG-157 No.2 Vol.2014-CVIM-194 No.2 2014/11/20. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report Viewpoint. 100. #154 (0,0,0). 50. z (mm). Projection Surface 0. Pixel -50. Observed Voxel. -150 -150. -100. -50. 0. 50. 100. x (mm). Depth. -100 #672 (-64.2,-1.0,-6.1). Filled Voxel. Similarity=90.1%. Cross-section of a 3-D occupancy grid. ਤ 2 ྨࣅ࢟੎ͷྫ. Fig. 2 Example of similar pose. 3.4 ྨࣅ࢟੎ͷ୳ࡧ. ਤ 3. 3-D occupancy grid ͷ࡞੒ํ๏. Fig. 3 Building method for a 3-D occupancy grid. Ξοϓ) ಈ࡞தʹଞͷ෺ମ͕‫ׯ‬ব͠ͳ͍͜ͱ΋ඞཁͱ͞Ε. ର৅෺ͷ‫ܗ‬ঢ়ʹΑͬͯ͸ɼਤ 2 ʹྫࣔ͢ΔΑ͏ʹɼҟͳ. ΔɽՃ͑ͯɼϚχϐϡϨʔλ͕଴‫࢟ػ‬੎͔Β೺࣋఺खલͷ. Δ 2 ࢟੎ؒͷ౤Өը૾͕ࣅ௨ͬͨ‫ܗ‬ঢ়ͱͳΔ͜ͱ͕͋Γɼ. Ξϓϩʔν։࢝఺·ͰɼͳΔ΂͘ແཧͷͳ͍ಈ࡞ͰҠಈՄ. ͜͜·Ͱʹड़΂ͨख๏ͷΈͰ͸ɼ͜ͷΑ͏ͳྨࣅ࢟੎ʹΑ. ೳͰ͋Δ͜ͱ΋ॏཁͱͳΔ [14][15]ɽ. Δ‫ॴہ‬ղΛഉআ͢Δ͜ͱ͕೉͍͠ɽ ͜ΕΛղܾ͢ΔͨΊɼຊ‫Ͱڀݚ‬͸ɼࣄલʹϞσϧͷγϧ. ຊ‫Ͱڀݚ‬͸ɼ1) ର৅෺ͷ೺࣋఺͸ CAD Ϟσϧ͔Β͋Β ͔͡Ί‫ʹࢉܭ‬ΑΓ‫ٻ‬ΊΒΕ͍ͯΔɼ2) ϦϑτΞοϓ‫ܦ‬࿏͸. Τοτ͔Βྨࣅ࢟੎σʔλϕʔεΛ࡞੒͠ɼ֤ྖҬͷҐஔ. ೺࣋఺͔ΒࣄલʹҰҙʹܾఆͰ͖Δɼ3) ଴‫࢟ػ‬੎͔ΒΞϓ. ࢟੎ͷਪఆ݁Ռʹ͍ͭͯσʔλϕʔε͔ΒείΞ͕‫ࡏݱ‬Α. ϩʔν։࢝఺·ͰͷϚχϐϡϨʔλͷ‫ܦ‬࿏͸ಈ࡞‫ܭ‬ըʹΑ. Γߴ͘ͳΔ࢟੎Λ‫͢ࡧݕ‬Δ͜ͱͱͨ͠ɽ. Γࣗಈతʹ‫͞ࢉܭ‬ΕΔɼͷ 3 ͭΛલఏ৚݅ͱͯ͠ɼ೺͓࣋. ྨࣅ࢟੎σʔλϕʔε͸ɼ೚ҙͷೋͭͷϞσϧ఺Λத৺. ΑͼϦϑτΞοϓ͕Մೳͳ఺͕ଟ͍΄Ͳ೺࣋ʹదͨ࢟͠੎. ͱͯ͠ɼͦΕͧΕͷ఺ͷओํ޲ϑϨʔϜ͕Ұக͢ΔΑ͏Ϟ. Ͱ͋ΔͱΈͳ͠ɼର৅෺ͷ೺࣋఺΁ͷΞϓϩʔν͓ΑͼϦ. σϧΛճసɼฏߦҠಈͯ͠ೋ࣍‫ݩ‬ฏ໘ʹ౤Ө͠ɼ࢟੎ͷγ. ϑτΞοϓ‫ܦ‬࿏্ͷো֐෺ۭؒΛ 3-D occupancy grid[16]. ϧΤοτ͕ 90%Ҏ্ॏෳ͢Δ΋ͷΛɼྨࣅ࢟੎ΛऔΓಘΔ. ͱ‫ݺ‬͹ΕΔख๏ʹΑΓ࡞੒ͯ͠ɼΞϓϩʔν͓ΑͼϦϑτ. Ϟσϧ఺ͱͯ͠ϦετΞοϓ͢Δ͜ͱͰ࡞੒ͨ͠ɽͨͩ͠ɼ. ΞοϓͷࡍʹϚχϐϡϨʔλ͓Αͼର৅෺͕‫ׯ‬ব͠ͳ͍೺. ҰͭͷϞσϧ఺ͷۙ๣ͷ఺͸૬‫͢ࣅྨʹޓ‬Δ࢟੎ʹͳΔ͜. ࣋఺ͷ਺Λ਺্͑͛Δ͜ͱͰείΞϦϯάΛߦ͏͜ͱͱ. ͱ͕ଟ͘ɼ·ͨϞσϧর߹࣌ʹۙ๣୳ࡧΛߦ͏ͨΊɼ2 ఺. ͨ͠ɽ. ؒͷ௚ઢ‫͕཭ڑ‬ᮢ஋ҎԼ·ͨ͸๏ઢํ޲ͷ಺ੵ͕ᮢ஋Ҏ্ ͷ఺ʹ͍ͭͯ͸ྨࣅ࢟੎୳ࡧΛߦΘͳ͍Α͏ʹ͍ͯ͠Δɽ. 4.1 3-D occupancy grid ͷ࡞੒. Ґஔ࢟੎ਪఆ‫ޙ‬ͷྨࣅ࢟੎‫Ͱࡧݕ‬͸ɼγʔυϖΞͷϞσ. 3-D occupancy grid ͸ɼࡾ࣍‫ߏࢠ֨ݩ‬଄Λ΋ͭ࡞‫ۭؒۀ‬. ϧ఺ΛΩʔͱͯ͠σʔλϕʔεΛ‫͠ࡧݕ‬ɼྨࣅ࢟੎ΛͱΔ. Λ༻ҙ͠ɼਤ 3 ʹࣔ͢Α͏ʹɼೋ࣍‫ݩ‬౤Өը૾ͷըૉ͝ͱ. ఺Λൃ‫ͨ͠ݟ‬৔߹ʹ͸‫ࡏݱ‬γʔυϖΞͷϞσϧ఺Λൃ‫͠ݟ‬. ʹɼ͋Β͔͡Ίઃఆͨ͠࠷ԕ఺͔Βըૉͷ࣋ͭࡾ࣍‫ݩ‬Ґஔ. ͨϞσϧ఺ͱஔ͖‫ͯ͑׵‬Ґஔ࢟੎ͷධՁΛߦ͍ɼධՁ஋ͷ. ·Ͱɼࢹ఺͔Βͷ‫཭ڑ‬ΛҰఆϐονͰ෼ׂͯ͠઎༗ۭؒΛ. ߴ͍ํͷϞσϧ఺Λ࠾༻͢Δ͜ͱΛ࠶‫ؼ‬తʹ‫܁‬Γฦ͢͜ͱ. ‫͠ࢉܭ‬ɼϘΫηϧΛੵΈ্͍͛ͯ͘͜ͱͰ࡞੒͢Δɽ. ͰɼେҬత࠷దղΛ‫ٻ‬ΊΔΑ͏ʹͨ͠ɽ. 4. ೺࣋Մೳੑʹ‫ͮ͘ج‬είΞϦϯά ϚχϐϡϨʔλʹΑΔ෦඼ͷ೺࣋Λ҆ఆͯ͠ߦ͏ͨΊʹ. ͜ΕʹΑΓ໘ͷΈͷߏ଄Ͱ͋Δ‫཭ڑ‬ը૾Λີͳߏ଄ʹม ‫͢׵‬Δ͜ͱͰɼ୯Ұࢹ఺͔ΒͳΔ͍ΘΏΔ 2.5 ࣍‫ߏݩ‬଄ͷ ‫཭ڑ‬ը૾Ͱ͸ෆՄೳͳɼࢹઢํ޲ʹରͯ͠ਫฏʹ͍ۙ໘Λ ࣋ͭ෺ମʹରͯ͠ͷ‫ׯ‬বνΣοΫ͕ՄೳͱͳΔɽ. ͸ɼର৅෺ͷҐஔ࢟੎ͷΈͳΒͣɼର৅෺ͷ‫ܗ‬ঢ়ͱ೺࣋Λ ߦ͏ΤϯυΤϑΣΫλͷ‫ܗ‬ঢ়ɼ೺࣋ํ๏ͷ૊Έ߹ΘͤʹΑ. 4.2 ϚχϐϡϨʔλͷ‫ׯ‬বνΣοΫ. Γܾఆ͞ΕΔɼ೺࣋ಈ࡞‫ޙ‬ͷର৅෺ͷ࢟੎͕҆ఆ͢Δʮ೺. ্ड़ͷํ๏Ͱ࡞੒ͨ͠࡞‫ۭؒۀ‬தͰΞϓϩʔν͓ΑͼϦ. ࣋఺ʯʹɼϚχϐϡϨʔλ͕઀ۙ (Ξϓϩʔν) Մೳͱͳ. ϑτΞοϓಈ࡞࣌ͷϚχϐϡϨʔλͷ‫ׯ‬বνΣοΫΛߦ. Δඞཁ͕͋Δɽ·ͨɼ೺࣋‫ޙ‬ͷର৅෺ͷ্࣋ͪ͛ (Ϧϑτ. ͏ͨΊʹɼຊ‫Ͱڀݚ‬͸ΤϯυΤϑΣΫλͷ‫ܗ‬ঢ়͓ΑͼΞϓ. ⓒ 2014 Information Processing Society of Japan. 3.

(4) Vol.2014-CG-157 No.2 Vol.2014-CVIM-194 No.2 2014/11/20. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report. ϩʔνɼϦϑτΞοϓಈ࡞ͷ‫੻ي‬Λԁ౵Ͱۙࣅ͢Δ͜ͱͱ ͨ͠ɽ ۙࣅԁ౵͸ϚχϐϡϨʔλͷखઌ࠲ඪ‫࡞Ͱܥ‬੒͞Εɼઌ ड़ͷํ๏Ͱਪఆ͞ΕͨϞσϧͷҐஔ࢟੎ຖʹɼ֤೺࣋఺ʹ ઃఆ͞Εͨ೺࣋࢟੎ʹ͋ΘͤͯճసɼฏߦҠಈΛߦ͍ɼΧ ϝϥ࠲ඪ‫࠲ʹܥ‬ඪม‫ ͯ͠׵‬3-D occupancy grid ಺ʹ഑ஔ ͞ΕΔɽ͜ͷ഑ஔ͞Εͨۙࣅԁ౵಺ʹϘΫηϧ͕ଘࡏ͢Δ ৔߹ɼϚχϐϡϨʔλʹ‫ׯ‬ব͢Δ෺ମ͕γʔϯதʹଘࡏ͢ Δͱ‫ͨ͠ͱͱ͜͢ͳݟ‬ɽ ·ͨɼΤϯυΤϑΣΫλ͕೺࣋ର৅෺ͱ઀͢Δ෦Ґ (ί. a) Pipes (found 5 objects). ϯλΫτΤϦΞ) Λۙࣅԁ౵಺ʹઃఆ͠ɼͦͷதʹϘΫη ϧ͕‫·ؚ‬Ε͍ͯͳ͍৔߹ʹ͸ɼҐஔ࢟੎ਪఆ͕‫͍ͯͬޡ‬Δ ΋ͷͱ‫͠ͳݟ‬ɼઃఆͨ͠఺਺Ҏ্ͷ೺࣋఺Ͱ‫ޡ‬ਪఆͱ൑ఆ ͞ΕͨҐஔ࢟੎ʹ͍ͭͯ͸೺࣋ީิ͔Βআ֎͢ΔΑ͏ʹ ͨ͠ɽ. 4.3 ର৅෺ͷ‫ׯ‬বνΣοΫ ‫ܗ‬ঢ়͕ൺֱతҰ༷ͳΤϯυΤϑΣΫλͱҟͳΓɼଟ༷ͳ ‫ܗ‬ঢ়Λ࣋ͭର৅෺͓Αͼͦͷ‫੻ي‬Λԁ౵Ͱۙࣅ͢Δ͜ͱ͸ ೉͍͠ɽຊ‫Ͱڀݚ‬͸ϦϑτΞοϓ࣌ͷ‫ׯ‬বνΣοΫʹ͍ͭ. b) Plates (found 2 objects). ͯɼ3-D occupancy grid ͔Βਪఆ͞ΕͨҐஔ࢟੎Λ࣋ͭର. ਤ 4. Ґஔ࢟੎ਪఆ݁Ռ. Fig. 4 Pose estimation results. ৅෺ΛऔΓআ͍ͨ࡞‫ۭؒۀ‬தͰɼϞσϧΛϦϑτΞοϓಈ ࡞ͷ‫͋ʹ੻ي‬ΘͤͯҰఆ෯ͰҠಈͤ͞ɼϞσϧΛߏ੒͢Δ ֤఺ͷҐஔͰͷ࡞‫ۭؒۀ‬தͷϘΫηϧͷ਺Λ਺্͑͛Δ͜. ‫͑׵‬ɼߏஙͨ͠σʔλϕʔε͔Βྨࣅ࢟੎Λ‫݁ͨ͠ࡧݕ‬Ռ. ͱͰߦ͏͜ͱͱͨ͠ɽ. Λࣔ͢ɽ. ͜ͷख๏͸ԁ౵ۙࣅʹൺֱͯ͠‫ࢉܭ‬ίετ͕େ͖͘ͳΔ. ྨࣅ࢟੎σʔλϕʔε͸ɼࢹઢํ޲͸ৗʹϞσϧ఺ͷ๏. ͨΊɼԁ౵ۙࣅʹΑΔϚχϐϡϨʔλͷ‫ׯ‬বνΣοΫʹ. ઢํ޲ͱҰக͍ͯ͠Δ΋ͷͱԾఆ͠ɼϞσϧ఺ؒͷ‫͕཭ڑ‬. ΑΓ೺࣋Մೳͱ͞Εͨ೺࣋఺ʹؔͯ͠ͷΈର৅෺ͷ‫ׯ‬ব. 30mm Ҏ্΋͘͠͸๏ઢํ޲ͷ֯౓͕ 30 ౓Ҏ্ҟͳΔ఺. νΣοΫΛ࣮ߦ͢Δ͜ͱͰɼ‫ؒ࣌ࢉܭ‬Λ࡟‫͢ݮ‬ΔΑ͏ʹ. ಉ࢜Ͱ࡞੒ͨ͠ɽ͜ͷ݁ՌϞσϧ఺ 1168 ‫ݸ‬ͷ͏ͪ 338 ૊. ͨ͠ɽ. ͷ૊Έ߹Θ͕ͤྨࣅ࢟੎ͱͳͬͨɽ. 5. ࣮‫݁ݧ‬Ռ͓Αͼߟ࡯ 5.1 Ґஔ࢟੎ਪఆ. Ϟσϧ఺ 633 Λ࠾༻ͨ͠‫࢟ޡ‬੎ (ਤ 5-a), ਤ 5-c) ͷ੺ͷ ఺) ͷ࢟੎ਪఆείΞ͸ 179 Ͱ͋ͬͨͷʹର͠ɼಉ఺ͷྨ ࣅ࢟੎‫݁ࡧݕ‬ՌͰ͋ΔϞσϧ఺ 655 Λ࠾༻ͨ͠ਖ਼͍࢟͠੎. ࠤ౻Βͷख๏ [17] ʹΑΓ࿪Έิਖ਼ࡁΈͷ PrimeSense ࣾ. (ਤ 5-b), ਤ 5-c) ͷ྘ͷ఺) ͷείΞ͸ 295 ͱͳͬͨɽྨࣅ. ੡ RGB-D Χϝϥ Xtion Λ༻͍ͯࡱ૾ͨ͠໿ 650mm ઌͷ. ࢟੎σʔλϕʔεதͰϞσϧ఺ 633 ͱ 655 ͷ࢟੎ྨࣅ౓. όϥੵΈঢ়ଶͷύΠϓ͓ΑͼԜತͷ͋Δ൘ঢ়෦඼ʹର͠. ͸ 92.3%ͱධՁ͞Ε͓ͯΓɼ࣮ࡍʹ྆࢟੎ؒͷ૬ҧ͸ࡾ࣍. ͯɼఏҊख๏Λద༻ͨ݁͠ՌΛਤ 4 ʹࣔ͢ɽೝࣝϞσϧ͸. ‫ݩ‬ը૾Λ༻͍ͯ΋ࢹೝ͢Δ͜ͱ͕೉͍͠ɽ͔͠͠ɼਤ 5 ͷ. STL ϑΥʔϚοτͷ CAD Ϟσϧ͔ΒྔࢠԽϐον 5mm. ന‫Ͱؙ‬ғΜͩ෦෼ʹ஫໨͢Ε͹ɼϞσϧ఺ 633 Λ࠾༻͠. Ͱ࡞੒͠ɼγʔϯ͸ 31x31pixel ͷ΢Οϯυ΢αΠζ (࣮ੇ. ͨਤ 5-a) ΑΓɼϞσϧ఺ 655 Λ࠾༻ͨ͠ਤ 5-b) ͷํ͕ਖ਼. Ͱ໿ 30mm ૬౰) Ͱಛ௃ྔͷ‫ࢉܭ‬Λߦͬͨɽ͜ͷ݁ՌΑΓ. ͍࢟͠੎Λ͍ࣔͯ͠Δ͜ͱ͕Θ͔Γɼྨࣅ࢟੎ʹ͓͍ͯ΋. ໌Β͔ͳΑ͏ʹɼຊख๏͸ɼର৅෺ͷ‫ܗ‬ঢ়ʹΑΒͣɼϊΠζ. Ϟσϧর߹ʹ͓͚Δ࢟੎ධՁ͕ਖ਼͘͠‫ػ‬ೳ͍ͯ͠Δ͜ͱ͕. ʹର͠ߴ͍‫ڧؤ‬ੑΛ࣋ͭͱ‫͖Ͱ͕ͱ͜͏ݴ‬Δɽ͜Ε͸γʔ. Θ͔Δɽ. ϯͷ΢Οϯυ΢ॲཧʹΑΔޮՌͷΈͳΒͣɼর߹ॲཧͰҐ. ͜ͷ͜ͱ͔Β໌Β͔ͳΑ͏ʹɼࣄલʹྨࣅ࢟੎σʔλ. ஔ࢟੎ͷީิͷબग़ɼධՁʹେྔͷγʔϯ఺Λ༻͍Δ͜ͱ. ϕʔεΛ࡞੒͠ɼ࢟੎ਪఆ݁ՌΛྨࣅ࢟੎ͱൺֱ͢Δख๏. ͰɼϥϯμϜϊΠζΛ౷‫ܭ‬తʹ཈͑ࠐΉͨΊͰ͋Δͱߟ͑. ͸ɼ‫ॴہ‬ղͱͯ͠ग़ྗ͞Εͨ‫࢟ͨͬޡ‬੎Λશ఺ͷর߹Λߦ. ΒΕΔɽ. ͏͜ͱͳ͠ʹमਖ਼͢ΔͨΊʹ༗ޮͰ͋Δͱ‫͑ݴ‬Δɽ. ͍࣍Ͱྨࣅ࢟੎୳ࡧͷ༗ޮੑΛ֬ೝ͢ΔͨΊɼਤ 5 ʹɼ ਤ 4-a) ͷதԝ෦ͷର৅෺ͷ࢟੎ਪఆ݁ՌΛਓҝతʹஔ͖. ⓒ 2014 Information Processing Society of Japan. ͨͩ͠ྨࣅ࢟੎‫ࡧݕ‬Λ༻͍ͯ΋ɼҐஔ࢟੎ͷਪఆ݁Ռʹ ύΠϓͷ௕࣠ํ޲ʹ࠷େ 10mm ఔ౓ɼਨ௚ํ޲ʹ࠷େ 5mm. 4.

(5) Vol.2014-CG-157 No.2 Vol.2014-CVIM-194 No.2 2014/11/20. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report. ਤ 6. a) Wrong pose (Model point633; Score=179). ਤ 4-a) ͷ 3-D occupancy grid ࡞੒݁Ռ. Fig. 6 3-D occupancy grid for Fig. 4-a) 57mm. 123mm. z Approach. y x. 75mm. 15mm. Contact Area 7mm. a) Manipulator occupancy model. b) Correct pose (Model point655; Score=295) Scene 520 540 560 580 600 620 640 660 680 -120 -100. -80. -80. -60. x axis in hand y axis z axis. 50 z (mm). -100. model grasp point. 100. Point #633 (Wrong). 520 540 560 580 600 620 640 660 680 -120. 0. -60. -40. -40. -20. -50. -20 0 20. -100. -50. 0. 50. 20. Point #655 (Collect). 520 540 560 580 600 620 640 660 680 -120. 0. 100. -100. -50. 50. 100. -100 150. Point #633 (Wrong) Point #655 (Collect). 520 540 560 580 600 620 640 660 680 -120. -100. 0. 100 50 0. y (mm). -50. -100. -80. -100. -80. -60. -150-100. -60. -40. -40. -20. 20. -100. -50. 0. 50. 100. 0 20. -100. -50. 0. 50. 100. c) 3-D view for scene and pose a), b) ਤ 5. -60. -40. -20. 0. 20. 40. 60. 80. 100. x (mm). b) Grasping points and poses. -20 0. -80. ྨࣅ࢟੎‫݁ࡧݕ‬Ռ. Fig. 5 Result of Pose-similarity search. ਤ 7 ϚχϐϡϨʔλϞσϧͱ೺࣋఺৘ใ. Fig. 7 Manipulator occupancy model and grasping points. ʹ 50mm ͷετϩʔΫͰΞϓϩʔνΛߦ͏͜ͱΛࣔͯ͠ ͍Δɽ·ͨ೺࣋఺ʹ౸ୡ‫ޙ‬ɼ‫ٵ‬ணͷͨΊʹ 7mm ͷԡ͠ࠐ. ఔ౓ͷ‫͕ࠩޡ‬ੜ͡Δ͜ͱ͕͋ͬͨɽ͜Εʹ͍ͭͯ͸ɼຊใ. Έಈ࡞Λߦ͏͜ͱ͔Βɼ‫ٵ‬ணύουʹ૬౰͢Δ 2 ຊͷ൒‫ܘ‬. ͰͷఏҊख๏ͷ‫ޙ‬ஈʹ ICP[18][19] ͳͲͷਫ਼ີͳҐஔ͋Θ. 7.5mmɼ௕͞ 7mm ͷԁ౵ۭؒΛίϯλΫτΤϦΞͱͯ͠. ͤख๏Λಋೖ͢Δ͜ͱͰରॲՄೳͰ͋Δͱߟ͑ΒΕΔɽ. ઃఆ͍ͯ͠Δɽ ਤ 7-b) Ͱ͸ɼΤϯυΤϑΣΫλͱର৅෺ͷ‫ׯ‬ব͓Αͼ. 5.2 ೺࣋Մೳੑʹ‫ͮ͘ج‬είΞϦϯά. ϦϑτΞοϓಈ࡞࣌ͷର৅෺ͷ࢟੎ͷ҆ఆੑΛߟྀ͠ɼ೺. ਤ 6 ʹਤ 4-a) ͷ 3-D occupancy grid Λ 5mm ϐονͰ. ࣋఺ͱͯ͠ 70 ఺ͷҐஔͱΤϯυΤϑΣΫλͷ࢟੎͕‫ࢉܭ‬. ࡞੒ͨ݁͠ՌΛɼਤ 7 ʹ͸ϚχϐϡϨʔλϞσϧͱ೺࣋఺. ͞Ε͍ͯΔ [20]ɽ·ͨϦϑτΞοϓํ޲͸෺ମ͓Αͼ೺࣋. ৘ใΛͦΕͧΕࣔ͢ɽ. ఺ͷҐஔ࢟੎ʹؔ܎ͳ͘ɼશͯࢹઢํ޲ͱҰக͢Δ΋ͷͱ. ਤ 6 Ͱ͸ࢹઢํ޲ʹରͯ͠΄΅ฏߦͳΔശͷน໘ͳͲ͕ ୆‫ܗ‬ঢ়ʹิؒ͞Ε͍ͯΔ͜ͱ͕Θ͔Δɽ. ͍ͯ͠Δɽ ͜ΕΒͷ৘ใΛ༻͍ͯɼਤ 4-a) ͷਪఆ݁Ռʹର͠ɼ೺࣋. ਤ 7-a) ͸ɼްΈ 80mm ͷΞʔϜͷઌ୺Λ‫ͯ͠ͱ఺ݪ‬൒‫ܘ‬. Մೳੑʹ‫ͮ͘ج‬είΞϦϯάΛ͓͜ͳͬͨ݁ՌΛਤ 8 ʹࣔ. 7.5mm ͷ‫ٵ‬ணύου 2 ຊΛ࣋ͬͨްΈ 50mmɼ௚‫ ܘ‬75mm. ͢ɽͳ͓ਤதͷείΞ͸౸ୡՄೳͱ‫͞ͳݟ‬Εͨ೺࣋఺ͷ਺. ͷΤϯυΤϑΣΫλΛ૷ண͠ɼखઌ࠲ඪ‫ ܥ‬X ࣠ͱ‫޲ํٯ‬. Λද͢ɽ. ⓒ 2014 Information Processing Society of Japan. 5.

(6) Vol.2014-CG-157 No.2 Vol.2014-CVIM-194 No.2 2014/11/20. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report. [4]. 3:: sscore=1 3 core e 1 1: score score=5 5 [5]. [6]. 2: score=2. ਤ 8. [7]. ೺࣋Մೳੑʹ‫ͮ͘ج‬είΞϦϯά݁Ռ. Fig. 8 Matching results using graspability information. [8]. ਤ 4-a) Ͱ͸ 5 ‫ݸ‬ͷީิ͕ϦετΞοϓ͞Ε͍ͯΔ͕ɼ ਤ 8 Ͱ͸ 3 ‫͍ͯͬݮʹݸ‬Δɽ͜Ε͸ɼ࡟আ͞Εͨީิʹ ͍ͭͯɼΞϓϩʔν࣌ʹശͷଆ໘΍ଞͷର৅෺ʹϚχϐϡ. [9]. Ϩʔλ͕‫ׯ‬ব͠ɼશͯͷ೺࣋఺ʹ౸ୡͰ͖ͳ͔ͬͨͨΊͰ ͋Δɽ. [10]. ࣮ࡍͷ೺࣋ಈ࡞Λ઒ా޻‫ۀ‬੡ HiroNX Λ༻͍ͯߦͬͨͱ ͜Ζɼ෺ମͷҐஔ࢟੎͓Αͼ೺࣋Մೳͱਪఆ͞Εͨ೺࣋఺ ͷҐஔΛಈ࡞‫ܭ‬ըϓϩάϥϜ Choreonoid graspPlugin[21]. [11]. ʹ౉͢͜ͱͰɼྑ޷ͳ݁ՌΛಘΔ͜ͱ͕Ͱ͖ͨɽ Ҏ্ΑΓɼ෺ମͷҐஔ࢟੎ਪఆΛߦ͏ࡍʹ೺࣋ՄೳੑΛ. [12]. ߟྀ͢Δ͜ͱͰɼಈ࡞‫ܭ‬ըϓϩάϥϜʹೋ࣍‫ݩ‬ը૾΍‫཭ڑ‬ ը૾Λ౉͢ͳͲෳࡶͳ‫ڠ‬ௐ੍‫ޚ‬ΛߦΘͳ͘ͱ΋࣮༻తͳϏ. [13]. ϯϐοΩϯά͕ՄೳʹͳΔͱ͍͏͜ͱ͕‫͑ݴ‬Δɽ. 6. ͓ΘΓʹ. [14]. ຊใͰ͸ɼओํ޲ϕΫτϧͱ‫཰ۂ‬Λ༻͍ͨࡾ࣍‫ݩ‬Ϟσϧ র߹ख๏ʹՃ͑ɼྨࣅ࢟੎σʔλϕʔεͷ࡞੒͢Δ͜ͱͰɼ. [15]. ର৅෺ͷ‫ܗ‬ঢ়ʹΑΒͣγʔϯͷϊΠζ΍ܽଛʹରͯ͠‫ڧؤ‬ ͳҐஔ࢟੎ਪఆ͕ՄೳͱͳΔ͜ͱΛࣔͨ͠ɽ·ͨɼҐஔ࢟ ੎ਪఆͷࡍʹɼ೺࣋఺΁ͷϚχϐϡϨʔλͷ౸ୡՄೳੑΛ ߟྀͨ͠είΞϦϯάΛߦ͏͜ͱͰɼώϡʔϦεςΟοΫ. [16]. ͳॲཧ΍Ϗδϣϯͱಈ࡞‫ܭ‬ըϓϩάϥϜͷؒͰෳࡶͳ‫ڠ‬ௐ ੍‫ޚ‬Λߦ͏͜ͱͳ͠ʹɼ࣮༻తͳϏϯϐοΩϯά͕Մೳͱ. [17]. ͳΔ͜ͱΛࣔͨ͠ɽ ࠓ‫ޙ‬ɼఆྔతͳਫ਼౓ධՁ͓Αͼଞͷख๏ͱͷൺֱΛߦ͏ ͜ͱͰɼຊख๏ͷ࣮༻ԽΛਐΊ͍ͯ͘༧ఆͰ͋Δɽ. [18]. ࢀߟจ‫ݙ‬ [1]. [2]. [3]. Sumi, Y., Kawai, Y., Yoshimi, T. and Tomita, F.: 3D Object Recognition in Cluttered Environments by Segment-Based Stereo Vision, International Journal of Computer Vision, Vol. 46, pp. 5–23 (2002). ‫݈ࢁؙ‬ҰɼՏҪྑߒɼ෋ాจ໌ɿःณྠֲઢΛ༻͍ͨ෺ ମϞσϧʹ‫ ͮ͘ج‬3 ࣍‫ݩ‬෺ମҐஔ࢟੎ਪఆɼి‫ֶؾ‬ձ࿦ จࢽ DɼVol. 131, pp. 505–514 (2011). Feldmar, J. and Ayache, N.: Rigid, affine and locally affine registration of free-form surfaces, International journal of computer vision, Vol. 18, pp. 99–119 (1996).. ⓒ 2014 Information Processing Society of Japan. [19]. [20]. [21]. Johnson, A. E. and Martial, H.: Using spin images for efficient object recognition in cluttered 3D scenes, PAMI on IEEE, Vol. 21, pp. 433–449 (1999). Rusu, R. B., Blodow, N. and Beetz, M.: Fast point feature histograms (FPFH) for 3D registration, ICRA’09 on. IEEE, pp. 3212–3217 (2009). ळ݄लҰɼ‫ڮ‬ຊɹֶɿಛ௃త 3-D ϕΫτϧϖΞΛ༻͍ͨ ͹ΒੵΈ෦඼ͷߴ଎Ґஔ࢟੎ೝࣝɼి‫ֶؾ‬ձ࿦จࢽ Cɼ Vol. 133, pp. 1853–1854 (2013). C.Dupuis, D., Léonard, S., Baumann, M. A., Croft, E. A. and Littley, J. J.: Two-Fingered Grasp Planning for Randomized Bin-Picking, Robotics: Science and Systems Workshop-Robot Manipulation: Intelligence in Human Environments (2008). Harada, K., Nagata, K., Tsuji, T., Yamanobe, N., Nakamura, A. and Kawai, Y.: Probabilistic Approach for Object Bin Picking Approximated by Cylinders, 2013 IEEE International Conference on Robotics and Automation (ICRA), pp. 3727–3732 (2013). Ghita, O. and Whelan, P. F.: A bin picking system based on depth from defocus, Machine Vision and Applications, Vol. 13, pp. 234–244 (2003). Sanz, P. J., Requena, A., esta, J. M. I. and Pobil, A. P. D.: Grasping the not-so-obvious: vision-based object handling for industrial applications, Robotics & Automation Magazine, Vol. 12, pp. 44–52 (2005). ੢ɹ୎࿠ɼ٢‫ݟ‬ɹོɼ‫ݚాݪ‬հɼՏҪྑߒɿϊΠζʹ‫ڧؤ‬ ͳࡾ࣍‫ݩ‬Ϟσϧর߹ख๏ɼୈ 17 ճը૾ͷೝࣝɾཧղγϯ ϙδ΢Ϝ (MIRU2014)ɼpp. SS3–24 (2014). Koenderink, J. J. and van Doorn, A. J.: Surface shape and curvature scales, Image and vision computing, Vol. 10, pp. 557–564 (1992). Fischler, M. A. and Bolles, R. C.: Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography, Communications of the ACM, Vol. 24, pp. 381–395 (1981). Bohg, J., Morales, A., Asfour, T. and Kragic, D.: DataDriven Grasp Synthesis—A Survey, IEEE Trans. on Robotics and Automation, Vol. 30, pp. 289–301 (2014). Nagata, K., Miyasaka, T., Nenchev, D. N., Yamanobe, N., Maruyama, K., Kawabata, S. and Kawai, Y.: Picking up an Indicated Object in a Complex Environment, Proc. of 2010 IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS’10), pp. 2109–2116 (2010). Elfes, A.: Sonar-Based Real-World Mapping and Navigation, IEEE Journal of Robotics and Automation, Vol. 3, pp. 249–265 (1987). ࠤ౻༤ོɼ‫݈࢘ాؠ‬ɼӬ‫ݟ‬෢࢘ɼ஛಺‫ޒܒ‬ɿRGB-D Χ ϝϥ͔ΒಘΒΕΔ Depth σʔλͷ࿪Έิਖ਼ɼϏδϣϯ ٕज़ͷ࣮ར༻ϫʔΫγϣοϓ (ViEW2013) ߨԋ࿦จू (CD-ROM)ɼpp. IS1–F6 (2013). Chen, Y. and Medioni, G.: Object modelling by registration of multiple range images, Proceedings of IEEE International Conference on Robotics and Automation, pp. 2724–2729 (1991). Besl, P. J. and McKay, N. D.: Method for registration of 3-D shapes, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 14, No. 2, pp. 239–256 (1992). Harada, K., Kaneko, K. and Kanehiro, F.: Fast grasp planning for hand/arm systems based on convex model, Robotics and Automation, 2008. ICRA 2008. IEEE International Conference on, pp. 1162–1168 (2008). ௰ɹಙੜɼ‫ݚాݪ‬հɿgraspPlugin for Choreonoid, ೔ຊ ϩϘοτֶձࢽɼ Vol. 31, No. 03, pp. 232–235 (2013).. 6.

(7)