[{"data":1,"prerenderedAt":257},["ShallowReactive",2],{"method-sibley2010swf":3},{"method":4,"reference":50,"equipment":70,"figures":83,"results":84},{"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":27,"sensors":33,"platform":35,"estimator":38,"association":39,"timeModel":40,"deskew":41,"loopClosure":42,"globalOptimization":43,"mapRepresentation":44,"prior":45,"outputGeometry":46,"compute":47,"codeUrl":48,"codeLicense":41,"relatedVersions":49},"sibley2010swf","Sibley et al., 2010","Sliding window filter","Sliding window filter with application to planetary landing",2010,"classic","C03","estimation_framework_or_library","本文以延遲狀態邊際化（delayed state marginalization）提出滑動視窗濾波器（SWF），用於提升行星著陸時長距離立體視覺的地表結構估計精度。方法在 k 個位姿的視窗內以含 Huber 核的穩健 Gauss-Newton 同時最佳化位姿與地標，並以 Schur 補把最舊位姿與不再被觀測的地標邊際化為先驗資訊；視窗涵蓋全部時間時等同完整 BA，只保留一個時間步時等同 EKF 的時間更新，逐步邊際化時則成為固定時間的方法。作者在實驗室以兩台 Point Grey Flea 相機組成的立體相機朝貼有 HiRISE 影像的平面牆移動，以 1:10 比例模擬著陸，結果顯示 3 至 5 影格的視窗已接近批次解；作者另指出在 10 個影格內，SWF 的誤差比視覺里程計低約 76%。但過早邊際化會鎖住線性化誤差，且邊際化使迴圈閉合難以處理。","A sliding window filter based on delayed marginalization spans full bundle adjustment (window over all time) to the EKF (one time step); in laboratory-emulated Mars descent stereo experiments, 3 to 5 frame windows approach the batch solution, and the authors report that over 10 frames the SWF error is about 76% lower than visual odometry, while early marginalization risks divergence and loop closures are effectively precluded.","full_text_reviewed","peer_reviewed_published","background","not_reported",[20,21],"controlled_experiment","simulation",[23,24,25,26],"Constant-time operation with results close to bundle adjustment (abstract; Fig. 14a)","Scales from offline batch to fast online incremental solutions (abstract; Fig. 9)","3-, 4- and 5-frame windows come close to the batch solution in the moving convergence experiment with 150 tracked features (Sec. 4 Moving Convergence; Fig. 13)","Over 10 frames, SWF error about 76% lower than visual odometry and standard deviation about 3.5 times smaller, at the same computational complexity as VO (Sec. 4 Effect of Marginalization)",[28,29,30,31,32],"Short windows (the EKF as a 1-frame SWF, or a 2-frame SWF) marginalize too early, lock in linearization errors and can diverge; a 10-frame window avoided this (Sec. 4 Effect of Early Marginalization; Fig. 15)","Marginalization makes landmark re-observation difficult and effectively precludes loop closures (Sec. 4)","Validation used a flat laboratory wall at 1:10 scale with printed orbital imagery; the authors note an image-scale mismatch with real landing imagery (Fig. 11 caption)","Sparse (VSDF-style diagonal) marginalization transfers measurement information into the prior less effectively than full marginalization (Sec. 4 Sparse Marginalization; Fig. 17)","The patch-warping model assumes locally planar patches with normals along the first optical axis and would need extension for general use (Sec. 4 The Importance of Feature Patch Warping)",[34],"stereo",[36,37],"laboratory stereo rig (two Point Grey Flea cameras, about 10 cm baseline) on a tripod moved in 15 cm steps along a translation stage from 10 m to 1 m toward a planar wall covered with printed MRO HiRISE imagery, emulating the last 100 m to 10 m of Mars descent at 1:10 scale","simulation (run-time and state-size experiments with 0.5 px image noise)","sliding window filter: robust Gauss-Newton (Huber kernel, typically 4 to 10 iterations) over all measurements of a k-pose window with a kinematic process model and a prior information term; the oldest poses and landmarks without active support are marginalized by the Schur complement into the prior (k = 1 reproduces the first-order EKF time step, a window over all time equals full BA or full SLAM)","sum-of-absolute-differences patch matching of Harris corners, Lucas-Kanade subpixel refinement with projective patch warping (local plane with normal along the first optical axis), Moravec's rigid-consistency check (greedy maximal clique) against gross outliers, then Huber M-estimation over the whole window","discrete poses","not_applicable","not handled: the authors state that marginalization makes re-observing landmarks difficult, which effectively precludes loop closures (not an issue for descent and landing)","window-limited; equivalent to full BA if the window covers all time","sparse 3D point landmarks (tracked surface features) with a possibly dense prior information block created by marginalization","no external map or survey prior; the internal prior information matrix collects the information of marginalized poses and landmarks, and a kinematic compound-operation process model links consecutive poses","sparse 3D landmark positions on the target surface; map error evaluated as the shortest distance from each landmark to a plane fitted to the wall (plane from the batch solution with fiducials)","constant run time independent of frame number shown in simulation averaged over 20 runs (10- and 20-frame windows with about 132 and 172 state parameters; Fig. 14a); computing hardware and absolute timings are not stated in the text; per-iteration patch re-alignment is described as expensive",null,[],{"id":5,"kind":51,"shortName":7,"title":8,"authors":52,"year":9,"venue":56,"venueType":57,"publisher":58,"volumeIssuePages":59,"doi":60,"arxivId":48,"url":61,"firstPublicDate":62,"publicationStatus":16,"metadataStatus":63,"fulltextStatus":15,"era":10,"classicReason":64,"codeUrl":48,"cluster":11,"topics":65,"mdpi":66,"verification":67,"label":6,"fulltextRoute":68,"versionRead":69,"addedByCensus":66},"method",[53,54,55],"Gabe Sibley","Larry Matthies","Gaurav Sukhatme","Journal of Field Robotics","journal","Wiley","27(5):587-608","10.1002\u002Frob.20360","https:\u002F\u002Fapi.crossref.org\u002Fworks\u002F10.1002\u002Frob.20360","2010-08-03","metadata_verified","principle reused: sliding-window estimation with delayed marginalization, spanning EKF to full BA, is the template for fixed-lag and sliding-window back-ends.",[11],false,"corrected","NTU institutional (Chrome)","Version of record, Journal of Field Robotics 27(5):587-608 (Wiley, first published 2010-08-03), HTML full text on Wiley Online Library",[71,77],{"category":72,"model":73,"canonical":73,"role":74,"dataset":48,"specs":75,"locator":76},"stereo_camera","Point Grey Research Flea","method input","two Flea cameras forming one stereo rig; about 10 cm baseline; narrow-field-of-view lenses (about 25 deg); grayscale images 1024 x 768 px; calibrated and rectified with CAHVOR camera models","Sec. 4 (Experimental Results); Fig. 10(c)",{"category":78,"model":79,"canonical":79,"role":80,"dataset":48,"specs":81,"locator":82},"other","tripod on a linear translation stage slotted onto a floor-fixed measuring rule (model not reported)","reference or ground truth","stereo rig moved repeatedly to predetermined locations in 15 cm steps from 10 m to 1 m from the wall; sequences taken with and without fiducials on the wall","Sec. 4 (Experimental Results); Fig. 11(a)",[],{"totalRows":85,"groupCount":86,"groups":87,"others":256},9,4,[88,158,197,225],{"slug":89,"group":90,"sourceId":91,"sourceLabel":92,"table":93,"selfRows":94,"metrics":95,"seqs":107,"entrants":112,"cells":123,"outcomes":152,"locators":153,"hardware":154,"wordings":155,"notes":156},"msckf2-2013-table-3","msckf2_2013:Table 3","msckf2_2013","Li & Mourikis, 2013","Table 3",3,[96,100,103],{"label":97,"unit":98,"statistic":99,"alignment":18},"Position RMSE (m)","m","RMSE",{"label":101,"unit":102,"statistic":99,"alignment":18},"Orientation RMSE (deg)","deg",{"label":104,"unit":105,"statistic":106,"alignment":18},"NEES","unitless","mean",[108],{"dataset":109,"sequence":110,"environment":111},"simulation from Cheddar Gorge dataset","Cheddar Gorge (29 km)","simulated long-distance driving",[113,116,119,121],{"name":114,"methodId":5,"linkable":115,"proposed":66,"self":115},"FLS (information-form fixed-lag smoother based on Sibley et al. 2010)",true,{"name":117,"methodId":118,"linkable":115,"proposed":66,"self":66},"MSCKF","mourikis2007msckf",{"name":120,"methodId":48,"linkable":66,"proposed":115,"self":66},"MSCKF 2.0",{"name":122,"methodId":48,"linkable":66,"proposed":66,"self":66},"'Ideal' MSCKF",[124,128,131,134,136,138,140,142,144,146,148,150],[125,125,125,126,127,125,127,127,125],0,133.4,-1,[129,125,125,130,127,125,127,127,125],1,146.2,[132,125,125,133,127,125,127,127,125],2,97.7,[94,125,125,135,127,125,127,127,125],100.2,[125,129,125,137,127,125,127,127,125],2.83,[129,129,125,139,127,125,127,127,125],3.4,[132,129,125,141,127,125,127,127,125],2.21,[94,129,125,143,127,125,127,127,125],2.35,[125,132,125,145,127,125,127,127,125],50.97,[129,132,125,147,127,125,127,127,125],51.72,[132,132,125,149,127,125,127,127,125],6.53,[94,132,125,151,127,125,127,127,125],6.45,[],[93],[],[],[157],"Monte Carlo simulation (50 trials) generated from the Cheddar Gorge dataset (29 km, 56 min driving, Xsens IMU at 100 Hz, images at 20 Hz); all methods use sliding windows of the same length",{"slug":159,"group":160,"sourceId":91,"sourceLabel":92,"table":161,"selfRows":132,"metrics":162,"seqs":168,"entrants":173,"cells":178,"outcomes":191,"locators":192,"hardware":193,"wordings":194,"notes":195},"msckf2-2013-text-sec-9","msckf2_2013:Text Sec.9","Text Sec.9",[163,166],{"label":164,"unit":98,"statistic":165,"alignment":18},"Largest position error (about)","max",{"label":167,"unit":98,"statistic":165,"alignment":18},"Worst-case elevation (altitude) error",[169],{"dataset":170,"sequence":171,"environment":172},"own vehicle dataset (Riverside, CA)","21.5 km drive","urban driving",[174,175,176],{"name":120,"methodId":48,"linkable":66,"proposed":115,"self":66},{"name":117,"methodId":118,"linkable":115,"proposed":66,"self":66},{"name":177,"methodId":5,"linkable":115,"proposed":66,"self":115},"FLS",[179,181,183,185,187,189],[125,125,125,180,127,125,127,127,125],58,[129,125,125,182,127,125,127,127,125],230,[132,125,125,184,127,125,127,127,125],202,[125,129,125,186,127,125,127,127,125],26,[129,129,125,188,127,125,127,127,125],27,[132,129,125,190,127,125,127,127,125],33,[],[161],[],[],[196],"Real vehicle run in Riverside, CA: 37 min, about 21.5 km, Xsens MTi-G at 100 Hz, one camera of a Bumblebee2 at 20 Hz, GPS-INS ground truth",{"slug":198,"group":199,"sourceId":5,"sourceLabel":6,"table":200,"selfRows":132,"metrics":201,"seqs":206,"entrants":208,"cells":213,"outcomes":218,"locators":219,"hardware":221,"wordings":222,"notes":223},"sibley2010swf-text-fig-14-caption","sibley2010swf:Text Fig. 14 caption","Text Fig. 14 caption",[202],{"label":203,"unit":204,"statistic":106,"alignment":205},"average size of the state vector","parameters","none",[207],{"dataset":21,"sequence":18,"environment":21},[209,211],{"name":210,"methodId":5,"linkable":115,"proposed":115,"self":115},"10-frame SWF",{"name":212,"methodId":5,"linkable":115,"proposed":115,"self":115},"20-frame SWF",[214,216],[125,125,125,215,127,125,127,127,125],132,[129,125,125,217,127,125,127,127,125],172,[],[220],"Fig. 14 caption",[],[],[224],"Simulation averaged over 20 runs, 76 landmarks tracked, about 20 features per frame, feature tracks about 10 frames, 0.5 px image noise; run-time curves themselves are plot-only",{"slug":226,"group":227,"sourceId":5,"sourceLabel":6,"table":228,"selfRows":132,"metrics":229,"seqs":237,"entrants":241,"cells":244,"outcomes":249,"locators":250,"hardware":252,"wordings":253,"notes":254},"sibley2010swf-text-sec-4-effect-of-marginalization","sibley2010swf:Text Sec.4 Effect of Marginalization","Text Sec.4 Effect of Marginalization",[230,233],{"label":231,"unit":232,"statistic":18,"alignment":18},"SWF error reduction relative to VO over 10 frames","%",{"label":234,"unit":235,"statistic":236,"alignment":18},"factor by which the SWF reduces the error standard deviation relative to VO over 10 frames","ratio","std",[238],{"dataset":239,"sequence":240,"environment":18},"not stated in the text (Sec. 4 Effect of Marginalization; the paragraph refers to Fig. 12 and to Newman et al. 2009)","10 frames",[242],{"name":243,"methodId":5,"linkable":115,"proposed":115,"self":115},"SWF",[245,247],[125,125,125,246,127,125,127,127,125],76,[125,129,125,248,127,125,127,127,125],3.5,[],[251],"Sec. 4 (Effect of Marginalization)",[],[],[255],"SWF versus visual odometry (VO, equivalent to deleting instead of marginalizing with k = 1) over 10 frames; the data set behind this comparison is not identified in the text",[],1790510663774]