[{"data":1,"prerenderedAt":529},["ShallowReactive",2],{"method-nubert2021delora":3},{"method":4,"reference":61,"equipment":81,"figures":112,"results":113},{"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":23,"limitations":28,"sensors":34,"platform":38,"estimator":42,"association":43,"timeModel":44,"deskew":45,"loopClosure":46,"globalOptimization":46,"mapRepresentation":47,"prior":48,"outputGeometry":49,"compute":50,"codeUrl":51,"codeLicense":52,"relatedVersions":53},"nubert2021delora","Nubert et al., 2021","DeLORA","Self-supervised Learning of LiDAR Odometry for Robotic Applications",2021,"recent","C09","odometry","DeLORA 以自監督（self-supervised）方式訓練 LiDAR 里程計網路：推論時只輸入由原始掃描投影成的球面距離影像，網路直接輸出相鄰兩幀的相對位姿；訓練時以 KD-tree 在三維空間尋找對應點，計算點到平面與平面到平面的幾何損失，因此不需要真值位姿或標註資料。法向量在訓練前以主成分分析（PCA）離線計算，只隨預測旋轉而轉動，使梯度不必穿過法向量計算。作者在足式機器人 ANYmal（建物地下室長廊）、履帶機器人（DARPA SubT Urban Circuit 資料）與 KITTI 上測試，並把位姿接到 LOAM 建圖模組產生地圖。","A self-supervised LiDAR odometry network trained with point-to-plane and plane-to-plane losses on 3D nearest-neighbour correspondences, needing no ground-truth poses, and tested on legged, tracked and wheeled platforms.","full_text_reviewed","peer_reviewed_published","main_body","在 ETH Zürich 建物地下室的長廊（類隧道）以 ANYmal 足式機器人測試，並使用 DARPA SubT Urban Circuit 資料（Satsop Business Park 的核電廠設施）；後者有地圖真值，但比較只作定性呈現。未在營建工地測試，也沒有以獨立量測評估點雲幾何精度。",[20,21,22],"completed_building","infrastructure","public_benchmark",[24,25,26,27],"no labelled or ground-truth data needed for training (abstract, Sec. III)","runs in real time on a mobile-class CPU (Sec. IV-A)","test mission used an upside-down LiDAR mounting while training used an upright mounting, and still produced a consistent map (Sec. IV-A, Fig. 4)","trained on the SubT Alpha course and tested on the unseen Beta course (Sec. IV-B)",[29,30,31,32,33],"odometry-only errors on unseen KITTI sequences 09 and 10 (6.05% and 6.44%) are about 3.6 to 3.9 times those after LOAM scan-to-map refinement (1.54% and 1.78%) (derived from Table II; the authors state that refinement helps especially on the test set)","ANYmal quantitative evaluation is relative to LOAM, because no external ground truth was available (Sec. IV-A, Table I)","SubT evaluation against the ground-truth map is qualitative (Sec. IV-B, Fig. 3)","IMU not used; the authors list multi-modal integration as future work (Sec. V)","(inference) training data from each new sensor or domain is still needed; no construction-site test",[35,36,37],"Velodyne VLP-16 Puck Lite (ANYmal, Sec. IV-A)","Ouster OS1-64 (DARPA SubT Urban, Sec. IV-B)","KITTI odometry LiDAR (sensor model not named in the paper, Sec. IV-C)",[39,40,41],"legged (ANYmal)","tracked UGV (iRobot PackBot Explorer, DARPA SubT Urban Circuit dataset)","vehicle (KITTI)","self-supervised CNN (ResNet-like blocks) regressing relative 6-DoF pose (translation + quaternion) from two spherical range images; poses optionally passed to the LOAM mapping module (Sec. III-B, IV-A)","training only: 3D nearest-neighbour correspondences via KD-tree, with point-to-plane and plane-to-plane losses on PCA normals precomputed offline; inference uses the range image (x, y, z, range) only (Sec. III-C, III-D)","discrete poses (scan-to-scan)","not_reported","none","none inside the method; maps in the experiments are built by the LOAM mapping module fed with DeLORA poses (Sec. IV-A, IV-B)","none at inference; training needs unlabelled scans from the target sensor or domain, no ground-truth poses (abstract, Sec. III)","6-DoF relative poses; point-cloud maps only when combined with LOAM mapping (Figs. 3-4)","about 48 ms per prediction on an i7-8565U laptop CPU and 13 ms on a GeForce MX250 laptop GPU, with about 32,000 points, H = 16, W = 720 (Sec. IV-A)","https:\u002F\u002Fgithub.com\u002Fleggedrobotics\u002FDeLORA","BSD-3-Clause (LICENSE file, copyright 2021 Julian Nubert, Robotic Systems Lab, ETH Zurich)",[54,58],{"relation":55,"title":56,"doi_or_url":57},"preprint","Self-supervised Learning of LiDAR Odometry for Robotic Applications (arXiv v1 2020-11-10, v2 2021-06-25)","https:\u002F\u002Farxiv.org\u002Fabs\u002F2011.05418",{"relation":59,"title":60,"doi_or_url":51},"code_release","DeLORA: Self-supervised Deep LiDAR Odometry for Robotic Applications",{"id":5,"kind":62,"shortName":7,"title":8,"authors":63,"year":9,"venue":67,"venueType":68,"publisher":69,"volumeIssuePages":70,"doi":71,"arxivId":72,"url":57,"firstPublicDate":73,"publicationStatus":16,"metadataStatus":74,"fulltextStatus":15,"era":10,"classicReason":75,"codeUrl":51,"cluster":11,"topics":76,"mdpi":77,"verification":78,"label":6,"fulltextRoute":79,"versionRead":80,"addedByCensus":77},"method",[64,65,66],"Julian Nubert","Shehryar Khattak","Marco Hutter","2021 IEEE International Conference on Robotics and Automation (ICRA), Xi'an, China","conference","IEEE","pp. 9601-9607","10.1109\u002Ficra48506.2021.9561063","2011.05418","2020-11-10","metadata_verified","not_applicable",[11],false,"corrected","arXiv","arXiv v2 (2021-06-25, 7 pages) read in full; IEEE Xplore version of record HTML full text (NTU institutional access, Chrome) compared section by section: body text of Sec. I to V identical to arXiv v2; its table images could not be loaded (IEEE media server returned 'temporarily unavailable'), so all table values come from arXiv v2",[82,90,95,101,104,109],{"category":83,"model":84,"canonical":85,"role":86,"dataset":87,"specs":88,"locator":89},"lidar","Velodyne VLP-16 Puck Lite","Velodyne VLP-16","method input",null,"mounted upright for the training missions and upside down for the test mission; range image H = 16, W = 720, about 32,000 points per scan","Sec. IV-A",{"category":91,"model":92,"canonical":92,"role":86,"dataset":87,"specs":93,"locator":94},"platform","ANYmal quadrupedal robot","learning-based locomotion controller; autonomous exploration missions of about 250 m on average in the ETH Zurich CLA basement","Sec. IV-A; Fig. 1",{"category":83,"model":96,"canonical":96,"role":97,"dataset":98,"specs":99,"locator":100},"Ouster OS1-64","dataset sensor","DARPA SubT Challenge Urban Circuit","carried by a tracked robot at Satsop Business Park; Alpha course used for training, Beta course for testing","Sec. IV-B",{"category":91,"model":102,"canonical":102,"role":97,"dataset":98,"specs":103,"locator":100},"iRobot PackBot Explorer (tracked robot)","tracked robot capable of fast in-spot yaw rotations",{"category":105,"model":106,"canonical":106,"role":107,"dataset":87,"specs":108,"locator":89},"compute","i7-8565U low-power laptop CPU","compute for runtime","about 48 ms per prediction",{"category":105,"model":110,"canonical":110,"role":107,"dataset":87,"specs":111,"locator":89},"GeForce MX250 laptop GPU","about 13 ms per prediction",[],{"totalRows":114,"groupCount":115,"groups":116,"others":507},44,8,[117,186,368,407],{"slug":118,"group":119,"sourceId":5,"sourceLabel":6,"table":120,"selfRows":121,"metrics":122,"seqs":129,"entrants":144,"cells":148,"outcomes":180,"locators":181,"hardware":182,"wordings":183,"notes":184},"nubert2021delora-table-i","nubert2021delora:Table I","Table I",12,[123,126],{"label":124,"unit":125,"statistic":45,"alignment":45},"t_rel [%] relative translation deviation vs LOAM","%",{"label":127,"unit":128,"statistic":45,"alignment":45},"r_rel [deg\u002F10m] relative rotation deviation vs LOAM","deg\u002F10m",[130,134,136,138,140,142],{"dataset":131,"sequence":132,"environment":133},"ANYmal CLA basement (own data)","test mission, segment length 5 m","indoor building basement with long tunnel-like corridors",{"dataset":131,"sequence":135,"environment":133},"test mission, segment length 10 m",{"dataset":131,"sequence":137,"environment":133},"test mission, segment length 25 m",{"dataset":131,"sequence":139,"environment":133},"test mission, segment length 40 m",{"dataset":131,"sequence":141,"environment":133},"test mission, segment length 60 m",{"dataset":131,"sequence":143,"environment":133},"test mission, segment length 100 m",[145],{"name":146,"methodId":5,"linkable":147,"proposed":147,"self":147},"Ours with mapping (DeLORA + LOAM mapping module)",true,[149,153,156,158,160,163,165,168,170,173,175,178],[150,150,150,151,152,150,152,152,150],0,0.345,-1,[150,154,150,155,152,150,152,152,150],1,0.484,[150,150,154,157,152,150,152,152,150],0.212,[150,154,154,159,152,150,152,152,150],0.274,[150,150,161,162,152,150,152,152,150],2,0.151,[150,154,161,164,152,150,152,152,150],0.15,[150,150,166,167,152,150,152,152,150],3,0.16,[150,154,166,169,152,150,152,152,150],0.103,[150,150,171,172,152,150,152,152,150],4,0.178,[150,154,171,174,152,150,152,152,150],0.069,[150,150,176,177,152,150,152,152,150],5,0.128,[150,154,176,179,152,150,152,152,150],0.046,[],[120],[],[],[185],"ANYmal test mission (LiDAR mounted upside down); relative pose deviation of DeLORA poses combined with the LOAM mapping module against the open-source LOAM implementation, which serves as reference because no external ground truth was available",{"slug":187,"group":188,"sourceId":5,"sourceLabel":6,"table":189,"selfRows":121,"metrics":190,"seqs":197,"entrants":206,"cells":232,"outcomes":361,"locators":363,"hardware":364,"wordings":365,"notes":366},"nubert2021delora-table-ii","nubert2021delora:Table II","Table II",[191,194],{"label":192,"unit":125,"statistic":193,"alignment":46},"t_rel [%]","mean",{"label":195,"unit":196,"statistic":193,"alignment":46},"r_rel [deg\u002F100m]","deg\u002F100m",[198,202,204],{"dataset":199,"sequence":200,"environment":201},"KITTI odometry","Training 00-08 (mean)","outdoor urban driving (car)",{"dataset":199,"sequence":203,"environment":201},"09",{"dataset":199,"sequence":205,"environment":201},"10",[207,209,211,213,216,218,220,222,224,226,229],{"name":208,"methodId":5,"linkable":147,"proposed":147,"self":147},"Ours",{"name":210,"methodId":5,"linkable":147,"proposed":147,"self":147},"Ours+Map (LOAM scan-to-map refinement)",{"name":212,"methodId":87,"linkable":77,"proposed":77,"self":77},"DeepLO [24]",{"name":214,"methodId":215,"linkable":147,"proposed":77,"self":77},"LO-Net [21]","lonet2019",{"name":217,"methodId":87,"linkable":77,"proposed":77,"self":77},"Velas et al. [20]",{"name":219,"methodId":87,"linkable":77,"proposed":77,"self":77},"UnDeepVO [8]",{"name":221,"methodId":87,"linkable":77,"proposed":77,"self":77},"SfMLearner [22]",{"name":223,"methodId":87,"linkable":77,"proposed":77,"self":77},"Zhu et al. [23]",{"name":225,"methodId":215,"linkable":147,"proposed":77,"self":77},"LO-Net+Map",{"name":227,"methodId":228,"linkable":147,"proposed":77,"self":77},"SUMA [13]","suma2018",{"name":230,"methodId":231,"linkable":147,"proposed":77,"self":77},"LOAM [3]","loam2017_auro",[233,234,236,238,240,242,243,245,247,249,251,252,254,256,258,260,262,264,266,268,270,272,274,276,278,280,281,283,284,286,287,289,291,293,295,297,299,302,304,306,308,310,312,315,317,319,321,323,325,327,329,331,333,335,337,340,342,344,346,347,348,351,353,355,357,359],[150,150,150,166,152,150,152,152,150],[150,154,150,235,152,150,152,152,150],1.38,[150,150,154,237,152,150,152,152,150],6.05,[150,154,154,239,152,150,152,152,150],2.15,[150,150,161,241,152,150,152,152,150],6.44,[150,154,161,166,152,150,152,152,150],[154,150,150,244,152,150,152,152,150],1.78,[154,154,150,246,152,150,152,152,150],0.73,[154,150,154,248,152,150,152,152,150],1.54,[154,154,154,250,152,150,152,152,150],0.68,[154,150,161,244,152,150,152,152,150],[154,154,161,253,152,150,152,152,150],0.69,[161,150,150,255,152,150,152,152,150],3.68,[161,154,150,257,152,150,152,152,150],0.87,[161,150,154,259,152,150,152,152,150],4.87,[161,154,154,261,152,150,152,152,150],1.95,[161,150,161,263,152,150,152,152,150],5.02,[161,154,161,265,152,150,152,152,150],1.83,[166,150,150,267,152,150,152,152,150],1.27,[166,154,150,269,152,150,152,152,150],0.67,[166,150,154,271,152,150,152,152,150],1.37,[166,154,154,273,152,150,152,152,150],0.58,[166,150,161,275,152,150,152,152,150],1.8,[166,154,161,277,152,150,152,152,150],0.93,[171,150,150,279,152,150,152,152,150],2.94,[171,154,150,87,150,150,152,152,150],[171,150,154,282,152,150,152,152,150],4.94,[171,154,154,87,150,150,152,152,150],[171,150,161,285,152,150,152,152,150],3.27,[171,154,161,87,150,150,152,152,150],[176,150,150,288,152,150,152,152,150],4.54,[176,154,150,290,152,150,152,152,150],2.55,[176,150,154,292,152,150,152,152,150],7.01,[176,154,154,294,152,150,152,152,150],3.61,[176,150,161,296,152,150,152,152,150],10.63,[176,154,161,298,152,150,152,152,150],4.65,[300,150,150,301,152,150,152,152,150],6,28.52,[300,154,150,303,152,150,152,152,150],4.67,[300,150,154,305,152,150,152,152,150],18.77,[300,154,154,307,152,150,152,152,150],3.21,[300,150,161,309,152,150,152,152,150],14.33,[300,154,161,311,152,150,152,152,150],3.3,[313,150,150,314,152,150,152,152,150],7,5.72,[313,154,150,316,152,150,152,152,150],2.35,[313,150,154,318,152,150,152,152,150],8.84,[313,154,154,320,152,150,152,152,150],2.92,[313,150,161,322,152,150,152,152,150],6.65,[313,154,161,324,152,150,152,152,150],3.89,[115,150,150,326,152,150,152,152,150],0.81,[115,154,150,328,152,150,152,152,150],0.44,[115,150,154,330,152,150,152,152,150],0.77,[115,154,154,332,152,150,152,152,150],0.38,[115,150,161,334,152,150,152,152,150],0.92,[115,154,161,336,152,150,152,152,150],0.41,[338,150,150,339,152,150,152,152,150],9,3.06,[338,154,150,341,152,150,152,152,150],0.89,[338,150,154,343,152,150,152,152,150],1.9,[338,154,154,345,152,150,152,152,150],0.8,[338,150,161,275,152,150,152,152,150],[338,154,161,154,152,150,152,152,150],[349,150,150,350,152,150,152,152,150],10,1.26,[349,154,150,352,152,150,152,152,150],0.5,[349,150,154,354,152,150,152,152,150],1.2,[349,154,154,356,152,150,152,152,150],0.48,[349,150,161,358,152,150,152,152,150],1.51,[349,154,161,360,152,150,152,152,150],0.57,[362],"not_reported (NA in table)",[189],[],[],[367],"KITTI odometry, errors over all subsequences of 100 to 800 m; DeLORA trained self-supervised on 00-08, tested on 09 and 10; only the 00-08 mean of LO-Net and Velas et al. was adapted by the authors because those were trained on 00-06",{"slug":369,"group":370,"sourceId":5,"sourceLabel":6,"table":371,"selfRows":115,"metrics":372,"seqs":375,"entrants":380,"cells":385,"outcomes":401,"locators":402,"hardware":403,"wordings":404,"notes":405},"nubert2021delora-table-iii","nubert2021delora:Table III","Table III",[373,374],{"label":192,"unit":125,"statistic":193,"alignment":46},{"label":195,"unit":196,"statistic":193,"alignment":46},[376,378],{"dataset":199,"sequence":377,"environment":201},"Training 00-06",{"dataset":199,"sequence":379,"environment":201},"Test 07-10",[381,383],{"name":382,"methodId":5,"linkable":147,"proposed":147,"self":147},"DeLORA loss variant: p2pl + pl2pl",{"name":384,"methodId":5,"linkable":147,"proposed":147,"self":147},"DeLORA loss variant: p2pl",[386,388,390,392,394,396,398,400],[150,150,150,387,152,150,152,152,150],3.41,[150,154,150,389,152,150,152,152,150],1.44,[150,150,154,391,152,150,152,152,150],8.3,[150,154,154,393,152,150,152,152,150],3.45,[154,150,150,395,152,150,152,152,150],6.47,[154,154,150,397,152,150,152,152,150],2.72,[154,150,154,399,152,150,152,152,150],8.9,[154,154,154,171,152,150,152,152,150],[],[371],[],[],[406],"Loss ablation on KITTI, networks trained from scratch on 00-06 and tested on 07-10",{"slug":408,"group":409,"sourceId":410,"sourceLabel":411,"table":412,"selfRows":171,"metrics":413,"seqs":419,"entrants":432,"cells":449,"outcomes":500,"locators":502,"hardware":503,"wordings":504,"notes":505},"nerfloam2023-table-3","nerfloam2023:Table 3","nerfloam2023","Deng et al., 2023","Table 3",[414],{"label":415,"unit":416,"statistic":417,"alignment":418},"RMSE of ATE (SE(3) alignment)","not stated","RMSE","SE3",[420,424,426,430],{"dataset":421,"sequence":422,"environment":423},"MaiCity","Mai00","synthetic urban street (simulated 64-beam LiDAR)",{"dataset":421,"sequence":425,"environment":423},"Mai01",{"dataset":427,"sequence":428,"environment":429},"Newer College","NC","outdoor campus, hand-carried LiDAR",{"dataset":199,"sequence":431,"environment":201},"KT09",[433,436,439,442,444,446,448],{"name":434,"methodId":435,"linkable":147,"proposed":77,"self":77},"ICP [3] (point-to-point)","besl1992icp",{"name":437,"methodId":438,"linkable":147,"proposed":77,"self":77},"GICP [31]","segal2009gicp",{"name":440,"methodId":441,"linkable":147,"proposed":77,"self":77},"Puma [36]","vizzo2021puma",{"name":443,"methodId":228,"linkable":147,"proposed":77,"self":77},"SuMA [2]",{"name":445,"methodId":5,"linkable":147,"proposed":77,"self":147},"DeLORA [26]",{"name":447,"methodId":87,"linkable":77,"proposed":77,"self":77},"PWC-LONet [39]",{"name":208,"methodId":410,"linkable":147,"proposed":147,"self":77},[450,451,453,455,457,459,461,463,465,467,469,471,473,475,477,479,480,482,484,485,487,489,491,493,495,496,497,498],[150,150,150,343,152,150,152,152,150],[150,150,154,452,152,150,152,152,150],0.05,[150,150,161,454,152,150,152,152,150],15.84,[150,150,166,456,152,150,152,152,150],5.86,[154,150,150,458,152,150,152,152,150],1.24,[154,150,154,460,152,150,152,152,150],0.13,[154,150,161,462,152,150,152,152,150],1.02,[154,150,166,464,152,150,152,152,150],34.25,[161,150,150,466,152,150,152,152,150],0.25,[161,150,154,468,152,150,152,152,150],0.06,[161,150,161,470,152,150,152,152,150],0.39,[161,150,166,472,152,150,152,152,150],3.58,[166,150,150,474,152,150,152,152,150],2.01,[166,150,154,476,152,150,152,152,150],0.04,[166,150,161,478,152,150,152,152,150],1.22,[166,150,166,176,152,150,152,152,150],[171,150,150,481,152,150,152,152,150],57.57,[171,150,154,483,152,150,152,152,150],5.12,[171,150,161,87,150,150,152,152,150],[171,150,166,486,152,150,152,152,150],29.09,[176,150,150,488,152,150,152,152,150],3.28,[176,150,154,490,152,150,152,152,150],0.09,[176,150,161,492,152,150,152,152,150],15.78,[176,150,166,494,152,150,152,152,150],4.6,[300,150,150,267,152,150,152,152,150],[300,150,154,460,152,150,152,152,150],[300,150,161,164,152,150,152,152,150],[300,150,166,499,152,150,152,152,150],4.26,[501],"failed",[412],[],[],[506],"Odometry ATE RMSE with SE(3) alignment (Sec. 5.1); '-' means failed; unit not stated; DeLORA and PWC-LONet are pre-trained on KITTI",[508,513,518,524],{"group":509,"slug":510,"sourceLabel":411,"table":511,"selfRows":171,"datasets":512},"nerfloam2023:Table 5","nerfloam2023-table-5","Table 5",[199],{"group":514,"slug":515,"sourceLabel":6,"table":516,"selfRows":161,"datasets":517},"nubert2021delora:Text Sec. IV-A","nubert2021delora-text-sec-iv-a","Text Sec. IV-A",[131],{"group":519,"slug":520,"sourceLabel":521,"table":522,"selfRows":154,"datasets":523},"zhang2024_3dlidarslam_survey:Table 8","zhang2024-3dlidarslam-survey-table-8","Zhang et al., 2024a","Table 8",[199],{"group":525,"slug":526,"sourceLabel":521,"table":527,"selfRows":154,"datasets":528},"zhang2024_3dlidarslam_survey:Table 9","zhang2024-3dlidarslam-survey-table-9","Table 9",[199],1790510661189]