[{"data":1,"prerenderedAt":103},["ShallowReactive",2],{"method-dissanayake2001":3},{"method":4,"reference":55,"equipment":77,"figures":102,"results":52},{"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":36,"platform":40,"estimator":42,"association":43,"timeModel":44,"deskew":45,"loopClosure":46,"globalOptimization":47,"mapRepresentation":48,"prior":49,"outputGeometry":50,"compute":51,"codeUrl":52,"codeLicense":53,"relatedVersions":54},"dissanayake2001","Dissanayake et al., 2001","EKF-SLAM convergence","A solution to the simultaneous localization and map building (SLAM) problem",2001,"classic","C01","full_slam_with_global_correction","本文以與 Smith 等人相同的估計理論架構，證明線性高斯情形下 EKF-SLAM 的三項性質：相對地圖不確定性單調下降、極限時地標估計完全相關、而絕對誤差下限只由初始車輛不確定性決定。作者強調維持完整地圖共變異數（交互相關項）是收斂與一致性的必要條件，省略它會造成不一致與發散。文中以毫米波雷達與車輛實作，並用測量過的地標位置進行比對，同時指出運算與儲存量隨地標數平方成長。","Proves convergence properties of full-covariance EKF-SLAM in the linear-Gaussian case and demonstrates it with a radar-equipped vehicle against surveyed landmarks, while noting O(N^2) cost.","full_text_reviewed","peer_reviewed_published","background","未在營建場域測試；實驗場地為布設 10 個雷達反射器的戶外測試場，場內另有大型貨運車輛與鄰近建物（Sec. IV-A.4、IV-B.2）。作者在引言把採礦與營建等任務中的自主全地形車輛列為 SLAM 的潛在應用（Sec. I，作者動機；作者自存稿與版本紀錄第 229 頁皆有此句）。作者並明言在地下礦坑等難以偵測幾何特徵的環境中，點地標策略不可行（Sec. V），與隧道及地下工程的退化問題相關（推論）。",[20,21],"controlled_experiment","independent_reference",[23,24,25,26],"Proof that relative map uncertainty decreases monotonically and absolute accuracy reaches a lower bound set by initial vehicle uncertainty (abstract; Sec. III).","Experimental results compared with surveyed landmark locations (abstract; Sec. IV).","Actual vehicle errors stayed inside the filter's 95% confidence bounds, so the estimates were consistent, conservative and non-divergent (Sec. IV-B.1, Fig. 8).","Landmark standard deviations decreased monotonically and reached a common lower bound matching the initial vehicle uncertainty, as predicted by the theory (Sec. IV-B.2, Figs. 12 and 13).",[28,29,30,31,32,33,34,35],"Computation and storage grow as N^2 with the number of landmarks (Sec. V).","Point-landmark framework is stated to be infeasible where geometric features are hard to detect, e.g., an underground mine (Sec. V).","Implementation is relatively small scale; map management for large areas is left open (Sec. V).","Durrant-Whyte and Bailey (2006, Part I, footnote 2 in Sec. III.C of the author-prepared copy; Sec. IV.A adds that convergence and consistency can only be guaranteed in the linear case) note that these results are proved only for the linear Gaussian case.","Only about 30% of radar observations corresponded to identifiable point landmarks; a large freight vehicle and nearby buildings also produced returns, so landmark identification and data association were essential (Sec. IV-A.4, Fig. 5).","The filter assumed a constant-velocity vehicle model, which caused a jump in x error when the vehicle accelerated; richer models add computation (Sec. IV-B.1).","Landmark estimates showed some bias, within the roughly 0.1 m accuracy of the surveyed truth (Sec. IV-B.2).","Using only point landmarks is inefficient indoors because information such as ranges to walls is not used (Sec. V).",[37,38,39],"millimetre-wave radar (77 GHz FMCW, beam scanned 360 deg in azimuth)","drive-shaft encoders (vehicle speed)","LVDT on the steering rack (steering for vehicle heading)",[41],"conventional utility vehicle, driven manually at speeds up to 10 m\u002Fs","extended Kalman filter over vehicle pose and point landmarks (linear-Gaussian analysis for the proofs)","point landmarks from thresholded radar returns; Appendix 2 maintains confirmed and tentative landmark lists and associates an observation to a landmark when its Mahalanobis-type distance is below a threshold (d_min); new landmarks are validated by observation counts and a quality measure","discrete poses","not_reported","implicit through full landmark covariance; no separate loop-closure module","none (recursive filter)","point landmark map with full covariance","none for estimation; surveyed landmark positions used only for evaluation","landmark coordinates and vehicle trajectory estimates with covariance","not_reported (full text read: radar, encoder and steering data were logged by an on-board computer and processed without reported hardware or timing)",null,"not_applicable",[],{"id":5,"kind":56,"shortName":7,"title":8,"authors":57,"year":9,"venue":63,"venueType":64,"publisher":65,"volumeIssuePages":66,"doi":67,"arxivId":52,"url":68,"firstPublicDate":69,"publicationStatus":16,"metadataStatus":70,"fulltextStatus":15,"era":10,"classicReason":71,"codeUrl":52,"cluster":11,"topics":72,"mdpi":73,"verification":74,"label":6,"fulltextRoute":75,"versionRead":76,"addedByCensus":73},"method",[58,59,60,61,62],"M.W.M.G. Dissanayake","P. Newman","S. Clark","H.F. Durrant-Whyte","M. Csorba","IEEE Transactions on Robotics and Automation","journal","IEEE","17(3):229-241","10.1109\u002F70.938381","https:\u002F\u002Fwww.robots.ox.ac.uk\u002F~mobile\u002FPapers\u002FSLAM_TransRandA.pdf","2001-06","metadata_verified","principle reused: gives the convergence properties of the full-covariance EKF-SLAM formulation (monotone decrease of map uncertainty, full correlation in the limit, lower bound set by initial vehicle uncertainty) that frame later consistency and drift discussions.",[11],false,"corrected","author copy","Author-hosted manuscript copy (robots.ox.ac.uk, Ghostscript PDF, 14 pp.) read in full, including Appendices 1 and 2; cross-checked against the IEEE T-RA version of record (arnumber 938381, 17(3):229-241) fetched through NTU access. Sec. IV and V wording and numbers match (77 GHz FMCW radar, 360 deg scan, 250 m range, 10 cm and 1.5 deg resolution, 30% of observations, speeds up to 10 m\u002Fs, about 5 cm reference path accuracy, 10 radar reflectors, bias within the true-measurement accuracy, underground-mine and 'relatively small scale' statements). Verifier cross-check (2026-09-25): PyMuPDF extracts the VoR text layer from page 1 as well, and the Sec. I mining and construction sentence is on VoR p. 229. The VoR uses American spelling and numbers subsections such as IV-B-1; most equations are images without a text layer.",[78,84,88,92,97],{"category":79,"model":80,"canonical":80,"role":81,"dataset":52,"specs":82,"locator":83},"radar","77 GHz FMCW millimetre-wave radar (MMWR)","method input","beam scanned 360 deg in azimuth at 1 to 3 Hz; amplitude returns at about 1.5 deg angular increments, thresholded to range and bearing; range to 250 m with 10 cm range and 1.5 deg bearing resolution; dual-polarisation receiver","Sec. IV-A",{"category":85,"model":86,"canonical":86,"role":81,"dataset":52,"specs":87,"locator":83},"wheel_or_leg_odometry","drive-shaft encoders","measure vehicle speed",{"category":89,"model":90,"canonical":90,"role":81,"dataset":52,"specs":91,"locator":83},"other","Linear Variable Differential Transformer (LVDT) on the steering rack","measures steering for vehicle heading",{"category":93,"model":94,"canonical":94,"role":81,"dataset":52,"specs":95,"locator":96},"platform","conventional utility vehicle","standard road vehicle fitted with the MMWR as primary sensor; driven manually, stationary about 30 s then loops at up to 10 m\u002Fs","Sec. IV-A, IV-B, Fig. 2",{"category":89,"model":98,"canonical":98,"role":99,"dataset":52,"specs":100,"locator":101},"radar reflectors (10, surveyed)","reference or ground truth","omni-directional point landmarks whose locations were accurately surveyed; used to evaluate the map and to compute the reference vehicle path (about 5 cm absolute accuracy)","Sec. IV-A, IV-B",[],1790510662638]