[{"data":1,"prerenderedAt":500},["ShallowReactive",2],{"method-kimeramulti2022":3},{"method":4,"reference":60,"equipment":83,"figures":105,"results":106},{"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":25,"sensors":32,"platform":36,"estimator":40,"association":41,"timeModel":42,"deskew":43,"loopClosure":44,"globalOptimization":45,"mapRepresentation":46,"prior":47,"outputGeometry":48,"compute":49,"codeUrl":50,"codeLicense":51,"relatedVersions":52},"kimeramulti2022","Tian et al., 2022","Kimera-Multi","Kimera-Multi: Robust, Distributed, Dense Metric-Semantic SLAM for Multi-Robot Systems",2022,"recent","C06","full_slam_with_global_correction","Kimera-Multi 是分散式多機器人度量語意 SLAM：各機器人以 Kimera-VIO（雙目與 IMU）估計軌跡並建立語意網格；相遇時交換詞袋描述子並做幾何驗證取得跨機迴圈；以分散式漸進非凸（D-GNC）穩健位姿圖最佳化剔除感知混淆造成的錯誤迴圈；最後以變形圖（deformation graph）依最佳化軌跡校正網格。","Kimera-Multi combines per-robot visual-inertial metric-semantic mapping with distributed place recognition, outlier-robust distributed PGO (D-GNC) and deformation-graph mesh correction.","full_text_reviewed","peer_reviewed_published","background","not_reported（驗證場景為照片級模擬 Medfield、City、Camp，EuRoC 室內序列，以及 Medfield 州立醫院園區與 MIT Stata Center 周邊的戶外資料，Sec. VII；未含工地）。網格精度只在 EuRoC Vicon 室與模擬器以真值點雲或網格評估。其分散式穩健位姿圖與以變形圖校正網格的架構，可作為多台機器人分區掃描工地後合併地圖的參考（推論）。",[20,21],"simulation","public_benchmark",[23,24],"Rejects incorrect loops from perceptual aliasing while fully distributed (abstract)","Estimation errors comparable to centralized SLAM with low bandwidth (abstract)",[26,27,28,29,30,31],"(inference) Visual-inertial only; no LiDAR geometry, so dense-mesh accuracy depends on stereo or RGB-D depth","With few inter-robot loop closures (Stata), default approximate D-GNC variable updates rejected the only inter-robot loop of robot 2 and misaligned its trajectory; correct alignment needed 2000 RBCD iterations (14 min) (Sec. VII-C, Fig. 14)","On Stata robot 1 the end-to-end error stayed large: 33.13 m for Kimera-Multi and 21.56 m for centralized GNC versus 24.19 m for Kimera-VIO (Table V)","Experiments assume robots are constantly within communication range; intermittent communication left to future work (Sec. VII-B)","Smaller communication savings on Machine Hall (five robots) because loose loop thresholds increase geometric-verification traffic (Sec. VII-B, Table II)","Mesh accuracy evaluated against ground truth only for EuRoC Vicon rooms and the simulator; outdoor datasets have end-to-end errors only (Sec. VII-B, VII-C)",[33,34,35],"stereo images and IMU (Kimera-VIO input)","depth images from an RGB-D camera or from stereo matching, plus 2D semantic segmentation (Kimera-Semantics input)","outdoor experiments: forward-facing RealSense D435i RGBD camera and IMU on a Clearpath Jackal UGV",[37,38,39],"wheeled UGV (Clearpath Jackal; Medfield and MIT Stata outdoor datasets, three robots each)","EuRoC sequences split into three or five simulated robots (platform not described in the paper)","photo-realistic simulation (DCIST Medfield, City, Camp)","per-robot Kimera-VIO; robust distributed pose-graph optimization via distributed graduated non-convexity (D-GNC) with RBCD solver","DBoW2 bag-of-words place recognition with geometric verification (five-point \u002F three-point RANSAC)","discrete poses (keyframes)","not_applicable","intra- and inter-robot visual loop closures with outlier rejection by D-GNC","distributed robust PGO; mesh corrected by deformation graph","metric-semantic 3D mesh","none","semantically labelled 3D mesh and optimized trajectories","Runs online on a CPU (Sec. III); CPU model not reported. Robust distributed PGO (D-GNC) took 8.9 s to 43.2 s, 2.2 s to 9.1 s with early stopping, versus 1.7 s to 6.3 s for centralized GNC (Table II). Outdoor: Medfield pose graph of 15650 poses solved with 100 RBCD iterations in 53 s; Stata (11184 poses) 120 iterations in 50 s, or 2000 iterations in 14 min for a correct alignment (Sec. VII-C). Communication 24.4 MB to 145.7 MB per dataset versus 1226 MB to 3685 MB for centralized image transfer (Table II).","https:\u002F\u002Fgithub.com\u002FMIT-SPARK\u002FKimera-Multi","not_verified",[53,57],{"relation":54,"title":55,"doi_or_url":56},"conference_version","Kimera-Multi: a System for Distributed Multi-Robot Metric-Semantic Simultaneous Localization and Mapping (ICRA 2021; Chang, Tian, How, Carlone)","10.1109\u002FICRA48506.2021.9561090",{"relation":58,"title":59,"doi_or_url":50},"code_release","MIT-SPARK\u002FKimera-Multi",{"id":5,"kind":61,"shortName":7,"title":8,"authors":62,"year":9,"venue":69,"venueType":70,"publisher":71,"volumeIssuePages":72,"doi":73,"arxivId":74,"url":75,"firstPublicDate":76,"publicationStatus":16,"metadataStatus":77,"fulltextStatus":15,"era":10,"classicReason":43,"codeUrl":50,"cluster":11,"topics":78,"mdpi":79,"verification":80,"label":6,"fulltextRoute":81,"versionRead":82,"addedByCensus":79},"method",[63,64,65,66,67,68],"Yulun Tian","Yun Chang","Fernando Herrera Arias","Carlos Nieto-Granda","Jonathan P. How","Luca Carlone","IEEE Transactions on Robotics","journal","IEEE","38(4):2022-2038","10.1109\u002Ftro.2021.3137751","2106.14386","https:\u002F\u002Farxiv.org\u002Fabs\u002F2106.14386","2021-06-28","metadata_verified",[11],false,"confirmed","arXiv","arXiv 2106.14386v2 (2021-12-17; accepted T-RO version); IEEE version of record not compared",[84,91,96,99],{"category":85,"model":86,"canonical":86,"role":87,"dataset":88,"specs":89,"locator":90},"platform","Clearpath Jackal UGV","method input","Medfield and Stata outdoor datasets (authors' own)","not_reported","Sec. VII-C",{"category":92,"model":93,"canonical":94,"role":87,"dataset":88,"specs":95,"locator":90},"rgbd","RealSense D435i RGBD Camera","Intel RealSense D435I","forward-facing",{"category":97,"model":98,"canonical":98,"role":87,"dataset":88,"specs":89,"locator":90},"imu","IMU (model not named; listed together with the RealSense D435i)",{"category":100,"model":101,"canonical":101,"role":102,"dataset":103,"specs":89,"locator":104},"stereo_camera","stereo camera and IMU of the EuRoC dataset (models not named in the paper)","dataset sensor","EuRoC (Vicon Room 1, Vicon Room 2, Machine Hall)","Sec. VII-B",[],{"totalRows":107,"groupCount":108,"groups":109,"others":499},44,4,[110,239,358,432],{"slug":111,"group":112,"sourceId":5,"sourceLabel":6,"table":113,"selfRows":114,"metrics":115,"seqs":122,"entrants":140,"cells":156,"outcomes":233,"locators":234,"hardware":235,"wordings":236,"notes":237},"kimeramulti2022-table-ii","kimeramulti2022:Table II","Table II",18,[116,119],{"label":117,"unit":118,"statistic":89,"alignment":47},"Communication [MB]","MB",{"label":120,"unit":121,"statistic":89,"alignment":47},"Runtime [sec] of robust PGO","s",[123,127,129,131,135,137],{"dataset":124,"sequence":125,"environment":126},"DCIST simulation","Medfield","photo-realistic simulation, 3 robots",{"dataset":124,"sequence":128,"environment":126},"City",{"dataset":124,"sequence":130,"environment":126},"Camp",{"dataset":132,"sequence":133,"environment":134},"EuRoC","Vicon Room 1","indoor room, 3 sequences as robots",{"dataset":132,"sequence":136,"environment":134},"Vicon Room 2",{"dataset":132,"sequence":138,"environment":139},"Machine Hall","industrial hall, 5 sequences as robots",[141,144,147,149,151,153],{"name":142,"methodId":5,"linkable":143,"proposed":143,"self":143},"Kimera-Multi total communication",true,{"name":145,"methodId":146,"linkable":79,"proposed":79,"self":79},"Centralized (Images)",null,{"name":148,"methodId":146,"linkable":79,"proposed":79,"self":79},"Centralized (Keypoints)",{"name":150,"methodId":5,"linkable":143,"proposed":143,"self":143},"D-GNC distributed",{"name":152,"methodId":5,"linkable":143,"proposed":143,"self":143},"D-GNC distributed (ES)",{"name":154,"methodId":155,"linkable":143,"proposed":79,"self":79},"Centralized GNC","yang2020gnc",[157,161,164,167,170,172,175,177,179,181,183,185,187,189,191,193,195,197,198,200,202,204,206,208,210,212,214,216,218,219,221,223,225,227,229,231],[158,158,158,159,160,158,160,160,158],0,65.9,-1,[162,158,158,163,160,158,160,160,158],1,2113,[165,158,158,166,160,158,160,160,158],2,141,[168,162,158,169,160,158,160,160,158],3,29.2,[108,162,158,171,160,158,160,160,158],5.9,[173,162,158,174,160,158,160,160,158],5,4.4,[158,158,162,176,160,158,160,160,158],69.5,[162,158,162,178,160,158,160,160,158],2326,[165,158,162,180,160,158,160,160,158],155,[168,162,162,182,160,158,160,160,158],22.1,[108,162,162,184,160,158,160,160,158],4.5,[173,162,162,186,160,158,160,160,158],3.2,[158,158,165,188,160,158,160,160,158],59.4,[162,158,165,190,160,158,160,160,158],3685,[165,158,165,192,160,158,160,160,158],246,[168,162,165,194,160,158,160,160,158],43.2,[108,162,165,196,160,158,160,160,158],9.1,[173,162,165,174,160,158,160,160,158],[158,158,168,199,160,158,160,160,158],27.8,[162,158,168,201,160,158,160,160,158],1226,[165,158,168,203,160,158,160,160,158],81.7,[168,162,168,205,160,158,160,160,158],8.9,[108,162,168,207,160,158,160,160,158],2.2,[173,162,168,209,160,158,160,160,158],3.1,[158,158,108,211,160,158,160,160,158],24.4,[162,158,108,213,160,158,160,160,158],1259,[165,158,108,215,160,158,160,160,158],83.9,[168,162,108,217,160,158,160,160,158],11.7,[108,162,108,186,160,158,160,160,158],[173,162,108,220,160,158,160,160,158],1.7,[158,158,173,222,160,158,160,160,158],145.7,[162,158,173,224,160,158,160,160,158],2362,[165,158,173,226,160,158,160,160,158],157,[168,162,173,228,160,158,160,160,158],20.5,[108,162,173,230,160,158,160,160,158],2.5,[173,162,173,232,160,158,160,160,158],6.3,[],[113],[],[],[238],"Communication usage (total of place recognition, geometric verification and distributed PGO) versus centralized baselines transmitting images or keypoints, and runtime of the robust PGO solver; hardware not reported",{"slug":240,"group":241,"sourceId":5,"sourceLabel":6,"table":242,"selfRows":243,"metrics":244,"seqs":248,"entrants":255,"cells":269,"outcomes":352,"locators":353,"hardware":354,"wordings":355,"notes":356},"kimeramulti2022-table-i","kimeramulti2022:Table I","Table I",12,[245],{"label":246,"unit":247,"statistic":89,"alignment":89},"Absolute trajectory error (ATE) [m]","m",[249,250,251,252,253,254],{"dataset":124,"sequence":125,"environment":126},{"dataset":124,"sequence":128,"environment":126},{"dataset":124,"sequence":130,"environment":126},{"dataset":132,"sequence":133,"environment":134},{"dataset":132,"sequence":136,"environment":134},{"dataset":132,"sequence":138,"environment":139},[256,258,260,262,264,266,268],{"name":257,"methodId":146,"linkable":79,"proposed":79,"self":79},"L2 (least squares, RBCD)",{"name":259,"methodId":146,"linkable":79,"proposed":79,"self":79},"PCM",{"name":261,"methodId":146,"linkable":79,"proposed":79,"self":79},"D-GNC (NI, naive initialization)",{"name":263,"methodId":146,"linkable":79,"proposed":79,"self":79},"PCM + D-GNC",{"name":265,"methodId":5,"linkable":143,"proposed":143,"self":143},"D-GNC",{"name":267,"methodId":5,"linkable":143,"proposed":143,"self":143},"D-GNC (ES, early stopping)",{"name":154,"methodId":155,"linkable":143,"proposed":79,"self":79},[270,272,274,276,278,280,282,285,287,289,291,293,295,297,298,300,302,304,306,308,310,312,314,315,317,319,321,323,325,327,329,331,333,335,337,339,341,343,344,346,348,350],[158,158,158,271,160,158,160,160,158],64.2,[162,158,158,273,160,158,160,160,158],12.5,[165,158,158,275,160,158,160,160,158],57.4,[168,158,158,277,160,158,160,160,158],4.64,[108,158,158,279,160,158,160,160,158],3.92,[173,158,158,281,160,158,160,160,158],4.32,[283,158,158,284,160,158,160,160,158],6,3.88,[158,158,162,286,160,158,160,160,158],3.58,[162,158,162,288,160,158,160,160,158],1.57,[165,158,162,290,160,158,160,160,158],0.91,[168,158,162,292,160,158,160,160,158],1.08,[108,158,162,294,160,158,160,160,158],0.85,[173,158,162,296,160,158,160,160,158],0.76,[283,158,162,162,160,158,160,160,158],[158,158,165,299,160,158,160,160,158],11.9,[162,158,165,301,160,158,160,160,158],1.37,[165,158,165,303,160,158,160,160,158],0.97,[168,158,165,305,160,158,160,160,158],1.09,[108,158,165,307,160,158,160,160,158],0.96,[173,158,165,309,160,158,160,160,158],0.75,[283,158,165,311,160,158,160,160,158],1.33,[158,158,168,313,160,158,160,160,158],1.17,[162,158,168,162,160,158,160,160,158],[165,158,168,316,160,158,160,160,158],0.34,[168,158,168,318,160,158,160,160,158],0.45,[108,158,168,320,160,158,160,160,158],0.35,[173,158,168,322,160,158,160,160,158],0.21,[283,158,168,324,160,158,160,160,158],0.36,[158,158,108,326,160,158,160,160,158],1.87,[162,158,108,328,160,158,160,160,158],1.56,[165,158,108,330,160,158,160,160,158],0.46,[168,158,108,332,160,158,160,160,158],0.62,[108,158,108,334,160,158,160,160,158],0.47,[173,158,108,336,160,158,160,160,158],0.48,[283,158,108,338,160,158,160,160,158],0.43,[158,158,173,340,160,158,160,160,158],1.92,[162,158,173,342,160,158,160,160,158],1.76,[165,158,173,336,160,158,160,160,158],[168,158,173,345,160,158,160,160,158],0.7,[108,158,173,347,160,158,160,160,158],0.41,[173,158,173,349,160,158,160,160,158],0.49,[283,158,173,351,160,158,160,160,158],0.52,[],[242],[],[],[357],"ATE in meters against ground truth for distributed trajectory estimators on Kimera-VIO odometry plus putative loops; fixed isotropic covariance (0.01 rad, 0.1 m); probability threshold 50%; statistic of the ATE not stated",{"slug":359,"group":360,"sourceId":5,"sourceLabel":6,"table":361,"selfRows":362,"metrics":363,"seqs":368,"entrants":386,"cells":392,"outcomes":426,"locators":427,"hardware":428,"wordings":429,"notes":430},"kimeramulti2022-table-iii","kimeramulti2022:Table III","Table III",8,[364],{"label":365,"unit":366,"statistic":89,"alignment":367},"Semantic label accuracy (%)","%","SE3",[369,372,374,376,378,380,382,384],{"dataset":124,"sequence":370,"environment":371},"Camp Robot 0","photo-realistic simulation",{"dataset":124,"sequence":373,"environment":371},"Camp Robot 1",{"dataset":124,"sequence":375,"environment":371},"Camp Robot 2",{"dataset":124,"sequence":377,"environment":371},"Camp Merged",{"dataset":124,"sequence":379,"environment":371},"City Robot 0",{"dataset":124,"sequence":381,"environment":371},"City Robot 1",{"dataset":124,"sequence":383,"environment":371},"City Robot 2",{"dataset":124,"sequence":385,"environment":371},"City Merged",[387,390],{"name":388,"methodId":389,"linkable":143,"proposed":79,"self":79},"Kimera-Semantics","kimera2020",{"name":391,"methodId":5,"linkable":143,"proposed":143,"self":143},"LMO (Kimera-Multi local mesh optimization)",[393,395,397,399,401,403,405,407,409,411,413,415,417,419,421,424],[158,158,158,394,160,158,160,160,158],81.6,[162,158,158,396,160,158,160,160,158],96.2,[158,158,162,398,160,158,160,160,158],92.8,[162,158,162,400,160,158,160,160,158],98.1,[158,158,165,402,160,158,160,160,158],82.8,[162,158,165,404,160,158,160,160,158],96.1,[158,158,168,406,160,158,160,160,158],79.4,[162,158,168,408,160,158,160,160,158],95.2,[158,158,108,410,160,158,160,160,158],77.1,[162,158,108,412,160,158,160,160,158],77.7,[158,158,173,414,160,158,160,160,158],80.7,[162,158,173,416,160,158,160,160,158],83.1,[158,158,283,418,160,158,160,160,158],71.4,[162,158,283,420,160,158,160,160,158],70.6,[158,158,422,423,160,158,160,160,158],7,76.1,[162,158,422,425,160,158,160,160,158],78.8,[],[361],[],[],[431],"Semantic label accuracy of meshes against the simulator's ground-truth labels after ICP registration (Open3D) of meshes sampled at 10^3 points per m^2, before (Kimera-Semantics) and after local mesh optimization (LMO)",{"slug":433,"group":434,"sourceId":5,"sourceLabel":6,"table":435,"selfRows":283,"metrics":436,"seqs":439,"entrants":455,"cells":461,"outcomes":493,"locators":494,"hardware":495,"wordings":496,"notes":497},"kimeramulti2022-table-v","kimeramulti2022:Table V","Table V",[437],{"label":438,"unit":247,"statistic":89,"alignment":47},"end-to-end error [m]",[440,444,446,448,451,453],{"dataset":441,"sequence":442,"environment":443},"Medfield outdoor dataset (authors' own)","Robot 0 (600 m)","outdoor campus with similar-looking scenes, Clearpath Jackal UGV",{"dataset":441,"sequence":445,"environment":443},"Robot 1 (860 m)",{"dataset":441,"sequence":447,"environment":443},"Robot 2 (728 m)",{"dataset":449,"sequence":450,"environment":443},"Stata outdoor dataset (authors' own)","Robot 0 (515 m)",{"dataset":449,"sequence":452,"environment":443},"Robot 1 (570 m)",{"dataset":449,"sequence":454,"environment":443},"Robot 2 (610 m)",[456,458,459],{"name":457,"methodId":389,"linkable":143,"proposed":79,"self":79},"Kimera-VIO",{"name":7,"methodId":5,"linkable":143,"proposed":143,"self":143},{"name":460,"methodId":155,"linkable":143,"proposed":79,"self":79},"Centralized",[462,464,466,467,469,471,472,474,476,477,479,481,482,484,486,488,490,492],[158,158,158,463,160,158,160,160,158],18.74,[162,158,158,465,160,158,160,160,158],0.01,[165,158,158,465,160,158,160,160,158],[158,158,162,468,160,158,160,160,158],14.84,[162,158,162,470,160,158,160,160,158],0.13,[165,158,162,470,160,158,160,160,158],[158,158,165,473,160,158,160,160,158],24.55,[162,158,165,475,160,158,160,160,158],0.09,[165,158,165,475,160,158,160,160,158],[158,158,168,478,160,158,160,160,158],49.02,[162,158,168,480,160,158,160,160,158],0.03,[165,158,168,465,160,158,160,160,158],[158,158,108,483,160,158,160,160,158],24.19,[162,158,108,485,160,158,160,160,158],33.13,[165,158,108,487,160,158,160,160,158],21.56,[158,158,173,489,160,158,160,160,158],29.35,[162,158,173,491,160,158,160,160,158],1.26,[165,158,173,313,160,158,160,160,158],[],[435],[],[],[498],"Outdoor datasets without ground truth: each robot starts and ends at the same place; end-to-end position error; Kimera-Multi uses D-GNC (Stata with full variable updates), centralized uses GNC in GTSAM",[],1790510661527]