[{"data":1,"prerenderedAt":370},["ShallowReactive",2],{"method-chebrolu2021adaptive":3},{"method":4,"reference":53,"equipment":74,"figures":87,"results":88},{"id":5,"label":6,"shortName":7,"title":8,"year":9,"era":10,"cluster":11,"scope":12,"keyIdeaZh":13,"keyIdeaEn":14,"fulltextStatus":15,"publicationStatus":16,"recommendation":17,"constructionRelevance":18,"validationEnvironment":19,"strengths":22,"limitations":29,"sensors":34,"platform":37,"estimator":40,"association":41,"timeModel":42,"deskew":42,"loopClosure":43,"globalOptimization":43,"mapRepresentation":42,"prior":43,"outputGeometry":44,"compute":45,"codeUrl":46,"codeLicense":47,"relatedVersions":48},"chebrolu2021adaptive","Chebrolu et al., 2021","Adaptive robust kernels","Adaptive Robust Kernels for Non-Linear Least Squares Problems",2021,"recent","C02","registration_component","作者以 Barron 的一般化穩健損失為基礎，把控制核形狀的參數 α 視為未知數，以交替最小化求解：先以一維格點搜尋在 [-10, 2] 內取殘差負對數概似最小的 α，再以迭代重加權最小平方法求解模型參數。為讓 α 可取負值以強力壓低離群值，作者把配分函數的積分截斷在 ±τ（τ = 10c），並預先建立解析度 0.1 的查找表。方法整合進 SuMa 的點對平面投影式 ICP，在 KITTI 里程計序列上不需人工離群值剔除即得到最低的平均平移誤差，並在 CARLA 模擬影像的光束法平差中擴大收斂範圍。","Uses a generalized robust kernel family whose shape is tuned automatically from residuals, tested in ICP and bundle adjustment.","full_text_reviewed","peer_reviewed_published","supplementary","not_reported",[20,21],"public_benchmark","simulation",[23,24,25,26,27,28],"best average KITTI relative translational error 2.03% vs 2.35% (Barron), 2.27% (Geman-McClure), 2.90% (SuMa hand-crafted rejection) and 6.34% (Huber) (Table I)","on sequence 04 Huber (49%) and the hand-crafted scheme (11.9%) fail while the adaptive kernel gives 0.95% (Table I, Sec. IV-A)","no hand-crafted outlier rejection needed; alpha becomes negative with moving cars on sequence 01 (Sec. IV-A, Fig. 5)","BA convergence rate 45% vs 24.8% (squared), 33% (Huber), 28.2% (Geman-McClure) (Sec. IV-B)","BA translation and rotation errors 2 to 5 times lower than with the Huber kernel depending on the dataset, as stated in the text (Sec. IV-B, Fig. 7)","results insensitive to tau in {10c, 20c, 50c, 100c}: at most about 5% translation and 8% rotation difference (Sec. IV-C)",[30,31,32,33],"not the best in relative rotational error (average 1.18 deg per 100 m vs 0.9 for the hand-crafted scheme and 0.92 for Huber) (Table I, Sec. IV-A)","scale parameter c is fixed; joint adaptation of alpha and c is unresolved (Sec. V)","a single alpha per scan pair or per BA iteration; per-object or per-block alpha is future work (Sec. V)","Agamennoni's elliptical-kernel method outperforms it in translation on the second (UAV) BA dataset (Sec. IV-B)",[35,36],"3D LiDAR scans of the KITTI odometry benchmark (scanner model not named in the paper)","monocular camera images simulated in CARLA (car front-looking, UAV nadir, strong shadows, side-looking with motion blur)",[38,39],"car (KITTI dataset)","simulation (CARLA: car-mounted and UAV cameras)","alternating minimization: (1) alpha chosen by 1-D grid search over [-10, 2] minimizing the negative log-likelihood of the current residuals under a truncated partition function (tau = 10c, lookup table at 0.1 resolution); (2) model parameters by IRLS with the Barron general kernel at that alpha; scale c fixed a priori","not part of the method; in the ICP experiment it runs inside SuMa frame-to-frame point-to-plane projective ICP without any outlier rejection step; in BA the initial matches come from SIFT with 5-point RANSAC","not_applicable","none","robust estimate (ICP pose or BA solution)","no runtime or hardware reported; the truncated partition function is precomputed as a lookup table for efficiency",null,"not_verified",[49],{"relation":50,"title":51,"doi_or_url":52},"preprint","arXiv:2004.14938 (v1 2020-04-30; v3 2021-02-18)","https:\u002F\u002Farxiv.org\u002Fabs\u002F2004.14938",{"id":5,"kind":54,"shortName":7,"title":8,"authors":55,"year":9,"venue":61,"venueType":62,"publisher":63,"volumeIssuePages":64,"doi":65,"arxivId":66,"url":52,"firstPublicDate":67,"publicationStatus":16,"metadataStatus":68,"fulltextStatus":15,"era":10,"classicReason":42,"codeUrl":46,"cluster":11,"topics":69,"mdpi":70,"verification":71,"label":6,"fulltextRoute":72,"versionRead":73,"addedByCensus":70},"component",[56,57,58,59,60],"Nived Chebrolu","Thomas Läbe","Olga Vysotska","Jens Behley","Cyrill Stachniss","IEEE Robotics and Automation Letters","journal","IEEE","6(2):2240-2247","10.1109\u002Flra.2021.3061331","2004.14938","2020-04-30","metadata_verified",[11],false,"confirmed","arXiv","arXiv 2004.14938v3 (2021-02-18; accepted version, marked as accepted for RA-L) read in full; IEEE RA-L version of record, vol. 6 no. 2 pp. 2240-2247 (April 2021), opened in Chrome under NTU access to check Table I on the rendered page (values identical to v3) and key text values (45%, 24.8%, 28.2%)",[75,81],{"category":76,"model":77,"canonical":77,"role":78,"dataset":79,"specs":18,"locator":80},"lidar","3D LiDAR (KITTI odometry scans; model not named in the paper)","dataset sensor","KITTI odometry","Sec. IV-A",{"category":82,"model":83,"canonical":83,"role":78,"dataset":84,"specs":85,"locator":86},"camera","simulated monocular camera (CARLA simulator)","authors' CARLA-simulated BA datasets","four authors' datasets: car front-looking, UAV nadir, strong shadows, side-looking with motion blur","Sec. IV-B, Fig. 6",[],{"totalRows":89,"groupCount":90,"groups":91,"others":369},25,2,[92,332],{"slug":93,"group":94,"sourceId":5,"sourceLabel":6,"table":95,"selfRows":96,"metrics":97,"seqs":104,"entrants":130,"cells":143,"outcomes":325,"locators":327,"hardware":328,"wordings":329,"notes":330},"chebrolu2021adaptive-table-i","chebrolu2021adaptive:Table I","Table I",24,[98,101],{"label":99,"unit":100,"statistic":18,"alignment":43},"relative rot. error in degrees per 100 m","deg\u002F100 m",{"label":102,"unit":103,"statistic":18,"alignment":43},"relative trans. error in %","%",[105,108,110,112,114,116,118,120,122,124,126,128],{"dataset":79,"sequence":106,"environment":107},"00","outdoor driving (KITTI)",{"dataset":79,"sequence":109,"environment":107},"01",{"dataset":79,"sequence":111,"environment":107},"02",{"dataset":79,"sequence":113,"environment":107},"03",{"dataset":79,"sequence":115,"environment":107},"04",{"dataset":79,"sequence":117,"environment":107},"05",{"dataset":79,"sequence":119,"environment":107},"06",{"dataset":79,"sequence":121,"environment":107},"07",{"dataset":79,"sequence":123,"environment":107},"08",{"dataset":79,"sequence":125,"environment":107},"09",{"dataset":79,"sequence":127,"environment":107},"10",{"dataset":79,"sequence":129,"environment":107},"Average",[131,134,136,138,140],{"name":132,"methodId":5,"linkable":133,"proposed":133,"self":133},"Our Approach",true,{"name":135,"methodId":46,"linkable":70,"proposed":70,"self":70},"Adaptive Kernel (Barron [6])",{"name":137,"methodId":46,"linkable":70,"proposed":70,"self":70},"Fixed Kernel (Huber)",{"name":139,"methodId":46,"linkable":70,"proposed":70,"self":70},"Fixed Kernel (Geman-McClure)",{"name":141,"methodId":142,"linkable":133,"proposed":70,"self":70},"Hand-Crafted Outlier Rejection [7] (SuMa original: Huber + rejection of correspondences >2 m or normal angle >30 deg)","suma2018",[144,148,151,153,155,157,159,161,163,166,168,171,173,176,178,181,183,185,187,190,192,194,195,198,200,202,203,205,207,209,210,211,212,214,215,216,217,218,219,221,222,223,224,226,227,228,229,231,233,234,235,236,238,240,242,244,245,246,248,249,250,252,253,254,255,257,259,261,262,263,264,266,268,269,271,272,273,274,275,276,277,278,279,281,282,284,285,286,287,288,289,290,291,292,293,295,297,299,300,301,302,303,304,305,306,307,309,310,311,313,314,315,316,317,318,319,320,321,322,323],[145,145,145,146,147,145,147,147,145],0,1.5,-1,[145,149,145,150,147,145,147,147,145],1,2.8,[145,145,149,152,147,145,147,147,145],1.3,[145,149,149,154,147,145,147,147,145],3.8,[145,145,90,156,147,145,147,147,145],0.91,[145,149,90,158,147,145,147,147,145],1.8,[145,145,160,146,147,145,147,147,145],3,[145,149,160,162,147,145,147,147,145],1.9,[145,145,164,165,147,145,147,147,145],4,0.81,[145,149,164,167,147,145,147,147,145],0.95,[145,145,169,170,147,145,147,147,145],5,0.97,[145,149,169,172,147,145,147,147,145],1.7,[145,145,174,175,147,145,147,147,145],6,0.51,[145,149,174,177,147,145,147,147,145],1.1,[145,145,179,180,147,145,147,147,145],7,2.1,[145,149,179,182,147,145,147,147,145],2.6,[145,145,184,152,147,145,147,147,145],8,[145,149,184,186,147,145,147,147,145],2.7,[145,145,188,189,147,145,147,147,145],9,0.8,[145,149,188,191,147,145,147,147,145],1.4,[145,145,193,152,147,145,147,147,145],10,[145,149,193,172,147,145,147,147,145],[145,145,196,197,147,145,147,147,145],11,1.18,[145,149,196,199,147,145,147,147,145],2.03,[149,145,145,201,147,145,147,147,145],1.6,[149,149,145,160,147,145,147,147,145],[149,145,149,204,147,145,147,147,145],1.2,[149,149,149,206,147,145,147,147,145],6.7,[149,145,90,208,147,145,147,147,145],0.93,[149,149,90,162,147,145,147,147,145],[149,145,160,191,147,145,147,147,145],[149,149,160,158,147,145,147,147,145],[149,145,164,213,147,145,147,147,145],0.82,[149,149,164,149,147,145,147,147,145],[149,145,169,170,147,145,147,147,145],[149,149,169,158,147,145,147,147,145],[149,145,174,175,147,145,147,147,145],[149,149,174,177,147,145,147,147,145],[149,145,179,220,147,145,147,147,145],2.2,[149,149,179,186,147,145,147,147,145],[149,145,184,152,147,145,147,147,145],[149,149,184,150,147,145,147,147,145],[149,145,188,225,147,145,147,147,145],0.88,[149,149,188,191,147,145,147,147,145],[149,145,193,204,147,145,147,147,145],[149,149,193,172,147,145,147,147,145],[149,145,196,230,147,145,147,147,145],1.19,[149,149,196,232,147,145,147,147,145],2.35,[90,145,145,208,147,145,147,147,145],[90,149,145,180,147,145,147,147,145],[90,145,149,204,147,145,147,147,145],[90,149,149,237,147,145,147,147,145],4.5,[90,145,90,239,147,145,147,147,145],0.79,[90,149,90,241,147,145,147,147,145],2.3,[90,145,160,243,147,145,147,147,145],0.7,[90,149,160,191,147,145,147,147,145],[90,145,164,177,145,145,147,147,145],[90,149,164,247,145,145,147,147,145],49,[90,145,169,239,147,145,147,147,145],[90,149,169,146,147,145,147,147,145],[90,145,174,251,147,145,147,147,145],0.64,[90,149,174,167,147,145,147,147,145],[90,145,179,204,147,145,147,147,145],[90,149,179,158,147,145,147,147,145],[90,145,184,256,147,145,147,147,145],0.96,[90,149,184,258,147,145,147,147,145],2.5,[90,145,188,260,147,145,147,147,145],0.78,[90,149,188,162,147,145,147,147,145],[90,145,193,170,147,145,147,147,145],[90,149,193,158,147,145,147,147,145],[90,145,196,265,147,145,147,147,145],0.92,[90,149,196,267,147,145,147,147,145],6.34,[160,145,145,158,147,145,147,147,145],[160,149,145,270,147,145,147,147,145],3.4,[160,145,149,152,147,145,147,147,145],[160,149,149,154,147,145,147,147,145],[160,145,90,149,147,145,147,147,145],[160,149,90,162,147,145,147,147,145],[160,145,160,146,147,145,147,147,145],[160,149,160,90,147,145,147,147,145],[160,145,164,225,147,145,147,147,145],[160,149,164,204,147,145,147,147,145],[160,145,169,280,147,145,147,147,145],0.98,[160,149,169,172,147,145,147,147,145],[160,145,174,283,147,145,147,147,145],0.62,[160,149,174,152,147,145,147,147,145],[160,145,179,182,147,145,147,147,145],[160,149,179,160,147,145,147,147,145],[160,145,184,146,147,145,147,147,145],[160,149,184,160,147,145,147,147,145],[160,145,188,149,147,145,147,147,145],[160,149,188,201,147,145,147,147,145],[160,145,193,152,147,145,147,147,145],[160,149,193,162,147,145,147,147,145],[160,145,196,294,147,145,147,147,145],1.32,[160,149,196,296,147,145,147,147,145],2.27,[164,145,145,298,147,145,147,147,145],0.9,[164,149,145,180,147,145,147,147,145],[164,145,149,204,147,145,147,147,145],[164,149,149,164,147,145,147,147,145],[164,145,90,189,147,145,147,147,145],[164,149,90,241,147,145,147,147,145],[164,145,160,243,147,145,147,147,145],[164,149,160,191,147,145,147,147,145],[164,145,164,177,145,145,147,147,145],[164,149,164,308,145,145,147,147,145],11.9,[164,145,169,189,147,145,147,147,145],[164,149,169,146,147,145,147,147,145],[164,145,174,312,147,145,147,147,145],0.6,[164,149,174,149,147,145,147,147,145],[164,145,179,204,147,145,147,147,145],[164,149,179,158,147,145,147,147,145],[164,145,184,149,147,145,147,147,145],[164,149,184,258,147,145,147,147,145],[164,145,188,189,147,145,147,147,145],[164,149,188,162,147,145,147,147,145],[164,145,193,149,147,145,147,147,145],[164,149,193,158,147,145,147,147,145],[164,145,196,298,147,145,147,147,145],[164,149,196,324,147,145,147,147,145],2.9,[326],"failed",[95],[],[],[331],"Frame-to-frame point-to-plane projective ICP inside SuMa on KITTI odometry; only the robust kernel or outlier scheme differs between rows",{"slug":333,"group":334,"sourceId":5,"sourceLabel":6,"table":335,"selfRows":149,"metrics":336,"seqs":339,"entrants":344,"cells":353,"outcomes":362,"locators":363,"hardware":365,"wordings":366,"notes":367},"chebrolu2021adaptive-text-sec-iv-b","chebrolu2021adaptive:Text Sec. IV-B","Text Sec. IV-B",[337],{"label":338,"unit":103,"statistic":18,"alignment":18},"successful convergence rate",[340],{"dataset":341,"sequence":342,"environment":343},"authors' CARLA-simulated bundle adjustment datasets","all 500 instances","simulation (CARLA)",[345,347,349,351],{"name":346,"methodId":5,"linkable":133,"proposed":133,"self":133},"Our approach (adaptive truncated kernel)",{"name":348,"methodId":46,"linkable":70,"proposed":70,"self":70},"squared loss",{"name":350,"methodId":46,"linkable":70,"proposed":70,"self":70},"Huber",{"name":352,"methodId":46,"linkable":70,"proposed":70,"self":70},"Geman-McClure",[354,356,358,360],[145,145,145,355,147,145,147,147,145],45,[149,145,145,357,147,145,147,147,145],24.8,[90,145,145,359,147,145,147,147,145],33,[160,145,145,361,147,145,147,147,145],28.2,[],[364],"Sec. IV-B (text)",[],[],[368],"BA convergence test: camera poses perturbed with sigma in [0.1 m, 5 m], 20 instances per noise level, 500 instances; converged if final camera-centre RMS error \u003C 1 cm",[],1790510662496]