[{"data":1,"prerenderedAt":452},["ShallowReactive",2],{"method-loner2023":3},{"method":4,"reference":56,"equipment":79,"figures":116,"results":156},{"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":35,"platform":37,"estimator":40,"association":41,"timeModel":42,"deskew":43,"loopClosure":44,"globalOptimization":45,"mapRepresentation":46,"prior":45,"outputGeometry":47,"compute":48,"codeUrl":49,"codeLicense":50,"relatedVersions":51},"loner2023","Isaacson et al., 2023","LONER","LONER: LiDAR Only Neural Representations for Real-Time SLAM",2023,"recent","C09","odometry_with_local_mapping","LONER 以點到平面 ICP（以單位矩陣為初始猜測，不使用 IMU）追蹤降採樣至 5 Hz 的 LiDAR 掃描，並在平行執行緒中以關鍵影格視窗（目前關鍵影格加上 7 個隨機選取的過去關鍵影格）聯合最佳化 MLP 與階層特徵格網地圖及關鍵影格位姿。其 JS 散度動態邊界損失依每條射線目前的學習程度調整目標分布寬度：尚未學好的區域用較大邊界以加快收斂，已學好的區域用較小邊界以保留細節。網格只在離線時以虛擬 LiDAR 與 marching cubes 產生；評估時將網格取樣為點雲、降採樣，並把真值裁切至感測器實際觀測範圍。","Real-time LiDAR-only SLAM with ICP tracking and a neural implicit map trained with an information-theoretic dynamic-margin loss.","full_text_reviewed","peer_reviewed_published","supplementary","論文未在營建場域測試；資料為 Fusion Portable（四足實驗室、手持校園庭院）與 Newer College Quad。",[20,21],"public_benchmark","simulation",[23,24,25,26],"Lower RMS APE than LeGO-LOAM on MCR, Canteen and Garden (0.029, 0.064, 0.056 m vs 0.052, 0.129, 0.161 m); on Newer College Quad LeGO-LOAM is 4 mm better (0.126 vs 0.130 m) (Table II)","Best completion on MCR and Canteen and best recall at 0.1 m on Canteen, Garden and Quad among the compared methods (Table III)","JS dynamic-margin loss converged faster than LOS and depth losses on a single simulated CARLA scan (Fig. 4), and the full loss gave the lowest or tied-lowest APE and L1 depth on every sequence in the loss ablation (Garden APE tied at 0.056 m with the LOS plus dynamic-margin row) (Table V)","Real-time operation: tracking 14 ms per step and each KeyFrame processed within its 3 s budget (Sec. IV-E)",[28,29,30,31,32,33,34],"No IMU; rapid rotation and feature-sparse scenes are future work (Sec. V)","Dynamic objects not handled (Sec. V)","City-scale operation not yet addressed (Sec. V)","Sky segmentation heuristic assumes the LiDAR is approximately level at initialization (Sec. III-E3)","SHINE Mapping, run with ground-truth poses, had better map accuracy on MCR and Quad and better completion on Garden and Quad than LONER (Table III)","Accuracy and precision were not reported for Canteen and Garden ('-', invalid configuration), so map quality there rests on completion and recall only (Table III)","Map metrics use voxel-downsampled clouds (5 cm, 1 cm for MCR) against ground truth cropped to observed geometry, so unobserved areas do not count against completeness (Sec. IV-D2)",[36],"3D LiDAR",[38,39],"handheld","legged","point-to-plane ICP tracking (identity initial guess, no IMU) + joint optimization of MLP map and keyframe poses in a window","point-to-plane ICP for tracking; depth-supervised neural rendering with a JS-divergence dynamic-margin loss for mapping","discrete poses","constant-velocity motion compensation between scans (Sec. III-B)","none reported","none","MLP with the hierarchical feature-grid encoding of ref. [26] (Instant-NGP multiresolution hash encoding) predicting volume density only, no colour (Sec. III-C)","mesh generated offline by virtual LiDAR rays + marching cubes (not part of online training)","AMD Ryzen 5950X CPU + NVIDIA A6000 GPU; tracking about 14 ms per step at the 5 Hz decimated scan rate; mapping 50 iterations per KeyFrame (one KeyFrame every 3 s) taking 2.79 s on average, about 56 ms per iteration or about 18 Hz map updates (Sec. IV-E)","https:\u002F\u002Fgithub.com\u002Fumautobots\u002FLONER","CC BY-NC-SA 4.0",[52],{"relation":53,"title":54,"doi_or_url":55},"preprint","arXiv:2309.04937","https:\u002F\u002Farxiv.org\u002Fabs\u002F2309.04937",{"id":5,"kind":57,"shortName":7,"title":8,"authors":58,"year":9,"venue":64,"venueType":65,"publisher":66,"volumeIssuePages":67,"doi":68,"arxivId":69,"url":70,"firstPublicDate":71,"publicationStatus":16,"metadataStatus":72,"fulltextStatus":15,"era":10,"classicReason":73,"codeUrl":49,"cluster":11,"topics":74,"mdpi":75,"verification":76,"label":6,"fulltextRoute":77,"versionRead":78,"addedByCensus":75},"method",[59,60,61,62,63],"Seth Isaacson","Pou-Chun Kung","Mani Ramanagopal","Ram Vasudevan","Katherine A. Skinner","IEEE Robotics and Automation Letters","journal","IEEE","8(12), 8042-8049","10.1109\u002Flra.2023.3324521","2309.04937","https:\u002F\u002Fapi.crossref.org\u002Fworks\u002F10.1109\u002Flra.2023.3324521","2023-09-10","metadata_verified","not_applicable",[11],false,"corrected","arXiv","arXiv v3 (2024-03-23), accepted manuscript marked 'IEEE RA-L preprint version, accepted October 2023'; IEEE version of record not read",[80,87,90,97,100,104,109],{"category":81,"model":82,"canonical":82,"role":83,"dataset":84,"specs":85,"locator":86},"compute","AMD Ryzen 5950X","compute for runtime",null,"CPU","Sec. IV-E",{"category":81,"model":88,"canonical":88,"role":83,"dataset":84,"specs":89,"locator":86},"NVidia A6000","GPU",{"category":91,"model":92,"canonical":92,"role":93,"dataset":94,"specs":95,"locator":96},"platform","quadruped (model not named)","dataset sensor","Fusion Portable","carried the sensors for MCR Slow 01, a small indoor lab scene","Sec. IV-C",{"category":91,"model":98,"canonical":98,"role":93,"dataset":94,"specs":99,"locator":96},"handheld platform (not named)","Canteen Day and Garden Day, medium-scale semi-outdoor courtyards",{"category":101,"model":102,"canonical":102,"role":93,"dataset":94,"specs":103,"locator":96},"stereo_camera","stereo camera (model not named)","used offline with RAFT optical flow to simulate RGB-D input for the NICE-SLAM baseline",{"category":105,"model":106,"canonical":106,"role":93,"dataset":107,"specs":108,"locator":96},"camera","monochrome fisheye cameras (model not named)","Newer College","reason NICE-SLAM could not be run on Newer College",{"category":110,"model":111,"canonical":111,"role":112,"dataset":113,"specs":114,"locator":115},"lidar","LiDAR (model not named)","method input","Fusion Portable; Newer College","only input sensor; scans decimated to 5 Hz; no IMU used (ICP starts from the identity)","Abstract; Sec. III-A, III-B; Sec. IV-C",[117,130,138,148],{"refId":5,"refLabel":6,"fig":118,"whatZh":119,"license":120,"licenseUrl":121,"sourceUrl":122,"src":123,"width":124,"height":125,"thumb":126,"thumbWidth":127,"thumbHeight":128,"modified":129},"Fig. 1","Fusion Portable 庭院場景的 LONER 網格重建與估計軌跡（紅線），周圍為訓練軌跡以外新視角渲染的深度影像","CC BY 4.0","https:\u002F\u002Fcreativecommons.org\u002Flicenses\u002Fby\u002F4.0\u002F","https:\u002F\u002Farxiv.org\u002Fhtml\u002F2309.04937v3\u002Ffigures\u002Ffig1_v4.png","\u002Ffigure-files\u002Floner2023\u002Ffig-1.webp",1400,900,"\u002Ffigure-files\u002Floner2023\u002Ffig-1.thumb.webp",480,309,"resized to at most 1400 px wide and converted to WebP",{"refId":5,"refLabel":6,"fig":131,"whatZh":132,"license":120,"licenseUrl":121,"sourceUrl":133,"src":134,"width":124,"height":135,"thumb":136,"thumbWidth":127,"thumbHeight":137,"modified":129},"Fig. 2","系統總覽：掃描降採樣後以 ICP 追蹤、分割天空、選取關鍵影格，並以新損失同時更新位姿與 MLP，地圖可離線渲染為深度影像或網格","https:\u002F\u002Farxiv.org\u002Fhtml\u002F2309.04937v3\u002Ffigures\u002Fsystem_overview_v10.png","\u002Ffigure-files\u002Floner2023\u002Ffig-2.webp",354,"\u002Ffigure-files\u002Floner2023\u002Ffig-2.thumb.webp",121,{"refId":5,"refLabel":6,"fig":139,"whatZh":140,"license":120,"licenseUrl":121,"sourceUrl":141,"src":142,"width":143,"height":144,"thumb":145,"thumbWidth":127,"thumbHeight":146,"modified":147},"Fig. 5","四個序列上 NICE-SLAM、SHINE、LONER 各損失版本的網格重建比較","https:\u002F\u002Farxiv.org\u002Fhtml\u002F2309.04937v3\u002Ffigures\u002Freconstruction_v9.png","\u002Ffigure-files\u002Floner2023\u002Ffig-5.webp",1024,383,"\u002Ffigure-files\u002Floner2023\u002Ffig-5.thumb.webp",180,"converted to WebP",{"refId":5,"refLabel":6,"fig":149,"whatZh":150,"license":120,"licenseUrl":121,"sourceUrl":151,"src":152,"width":124,"height":153,"thumb":154,"thumbWidth":127,"thumbHeight":155,"modified":129},"Fig. 6","以不同損失訓練的 MLP 所渲染的深度影像，比較補洞能力與幾何細節","https:\u002F\u002Farxiv.org\u002Fhtml\u002F2309.04937v3\u002Ffigures\u002Fdepth_image_for_loss_comparison_v2.png","\u002Ffigure-files\u002Floner2023\u002Ffig-6.webp",592,"\u002Ffigure-files\u002Floner2023\u002Ffig-6.thumb.webp",203,{"totalRows":157,"groupCount":158,"groups":159,"others":451},24,3,[160,349,414],{"slug":161,"group":162,"sourceId":5,"sourceLabel":6,"table":163,"selfRows":164,"metrics":165,"seqs":178,"entrants":190,"cells":203,"outcomes":341,"locators":344,"hardware":345,"wordings":346,"notes":347},"loner2023-table-iii","loner2023:Table III","Table III",16,[166,171,173,176],{"label":167,"unit":168,"statistic":169,"alignment":170},"Accuracy (mean distance estimated to ground truth)","m","mean","not_reported",{"label":172,"unit":168,"statistic":169,"alignment":170},"Completion (mean distance ground truth to estimated)",{"label":174,"unit":175,"statistic":170,"alignment":170},"Precision at 0.1 m threshold","ratio",{"label":177,"unit":175,"statistic":170,"alignment":170},"Recall at 0.1 m threshold",[179,182,185,187],{"dataset":94,"sequence":180,"environment":181},"MCR Slow 01","indoor lab, quadruped",{"dataset":94,"sequence":183,"environment":184},"Canteen Day","semi-outdoor courtyard, handheld",{"dataset":94,"sequence":186,"environment":184},"Garden Day",{"dataset":107,"sequence":188,"environment":189},"Quad Easy","large outdoor college quad, two laps",[191,195,198,200,202],{"name":192,"methodId":193,"linkable":194,"proposed":75,"self":75},"NICE-SLAM","niceslam2022",true,{"name":196,"methodId":197,"linkable":194,"proposed":75,"self":75},"SHINE (ground-truth poses)","shinemapping2023",{"name":199,"methodId":84,"linkable":75,"proposed":75,"self":75},"LONER w.\u002F L_CLONeR",{"name":201,"methodId":84,"linkable":75,"proposed":75,"self":75},"LONER w.\u002F L_URF",{"name":7,"methodId":5,"linkable":194,"proposed":194,"self":194},[204,208,211,214,216,219,221,223,225,227,229,231,233,235,237,239,241,243,245,247,249,250,251,252,253,254,255,257,259,261,263,264,265,266,267,268,269,271,273,275,277,278,279,280,281,282,283,285,287,289,291,292,293,294,295,296,297,299,301,303,305,306,308,310,312,314,315,317,319,321,323,324,326,328,330,332,333,335,337,339],[205,205,205,206,207,205,207,207,205],0,0.621,-1,[209,205,205,210,207,205,207,207,205],1,0.164,[212,205,205,213,207,205,207,207,205],2,0.11,[158,205,205,215,207,205,207,207,205],0.153,[217,205,205,218,207,205,207,207,205],4,0.186,[205,209,205,220,207,205,207,207,205],0.419,[209,209,205,222,207,205,207,207,205],0.075,[212,209,205,224,207,205,207,207,205],0.08,[158,209,205,226,207,205,207,207,205],0.102,[217,209,205,228,207,205,207,207,205],0.069,[205,212,205,230,207,205,207,207,205],0.124,[209,212,205,232,207,205,207,207,205],0.624,[212,212,205,234,207,205,207,207,205],0.665,[158,212,205,236,207,205,207,207,205],0.449,[217,212,205,238,207,205,207,207,205],0.473,[205,158,205,240,207,205,207,207,205],0.476,[209,158,205,242,207,205,207,207,205],0.757,[212,158,205,244,207,205,207,207,205],0.94,[158,158,205,246,207,205,207,207,205],0.884,[217,158,205,248,207,205,207,207,205],0.932,[205,205,209,84,205,205,207,207,205],[209,205,209,84,209,205,207,207,205],[212,205,209,84,209,205,207,207,205],[158,205,209,84,209,205,207,207,205],[217,205,209,84,209,205,207,207,205],[205,209,209,84,205,205,207,207,205],[209,209,209,256,207,205,207,207,205],0.116,[212,209,209,258,207,205,207,207,205],0.22,[158,209,209,260,207,205,207,207,205],0.19,[217,209,209,262,207,205,207,207,205],0.105,[205,212,209,84,205,205,207,207,205],[209,212,209,84,209,205,207,207,205],[212,212,209,84,209,205,207,207,205],[158,212,209,84,209,205,207,207,205],[217,212,209,84,209,205,207,207,205],[205,158,209,84,205,205,207,207,205],[209,158,209,270,207,205,207,207,205],0.753,[212,158,209,272,207,205,207,207,205],0.524,[158,158,209,274,207,205,207,207,205],0.846,[217,158,209,276,207,205,207,207,205],0.878,[205,205,212,84,205,205,207,207,205],[209,205,212,84,209,205,207,207,205],[212,205,212,84,209,205,207,207,205],[158,205,212,84,209,205,207,207,205],[217,205,212,84,209,205,207,207,205],[205,209,212,84,205,205,207,207,205],[209,209,212,284,207,205,207,207,205],0.13,[212,209,212,286,207,205,207,207,205],0.333,[158,209,212,288,207,205,207,207,205],0.539,[217,209,212,290,207,205,207,207,205],0.157,[205,212,212,84,205,205,207,207,205],[209,212,212,84,209,205,207,207,205],[212,212,212,84,209,205,207,207,205],[158,212,212,84,209,205,207,207,205],[217,212,212,84,209,205,207,207,205],[205,158,212,84,205,205,207,207,205],[209,158,212,298,207,205,207,207,205],0.657,[212,158,212,300,207,205,207,207,205],0.469,[158,158,212,302,207,205,207,207,205],0.623,[217,158,212,304,207,205,207,207,205],0.784,[205,205,158,84,209,205,207,207,205],[209,205,158,307,207,205,207,207,205],0.301,[212,205,158,309,207,205,207,207,205],0.663,[158,205,158,311,207,205,207,207,205],0.552,[217,205,158,313,207,205,207,207,205],0.38,[205,209,158,84,209,205,207,207,205],[209,209,158,316,207,205,207,207,205],0.148,[212,209,158,318,207,205,207,207,205],0.543,[158,209,158,320,207,205,207,207,205],0.895,[217,209,158,322,207,205,207,207,205],0.373,[205,212,158,84,209,205,207,207,205],[209,212,158,325,207,205,207,207,205],0.453,[212,212,158,327,207,205,207,207,205],0.15,[158,212,158,329,207,205,207,207,205],0.127,[217,212,158,331,207,205,207,207,205],0.327,[205,158,158,84,209,205,207,207,205],[209,158,158,334,207,205,207,207,205],0.717,[212,158,158,336,207,205,207,207,205],0.602,[158,158,158,338,207,205,207,207,205],0.484,[217,158,158,340,207,205,207,207,205],0.809,[342,343],"failed","not_applicable (invalid configuration '-')",[163],[],[],[348],"Map accuracy and completion (m, mean nearest-point distances) and precision and recall at a 0.1 m threshold; meshes from each method sampled to point clouds, all clouds voxel-downsampled to 5 cm (1 cm for MCR), ground truth cropped to geometry observed by the sensor. SHINE Mapping used ground-truth poses. '-' = invalid configuration (accuracy and precision are '-' for every method on Canteen and Garden; NICE-SLAM '-' on Quad), x = failed (NICE-SLAM on Canteen and Garden, all metrics).",{"slug":350,"group":351,"sourceId":5,"sourceLabel":6,"table":352,"selfRows":217,"metrics":353,"seqs":357,"entrants":362,"cells":370,"outcomes":407,"locators":409,"hardware":410,"wordings":411,"notes":412},"loner2023-table-ii","loner2023:Table II","Table II",[354],{"label":355,"unit":168,"statistic":356,"alignment":170},"RMS APE (t_APE)","RMSE",[358,359,360,361],{"dataset":94,"sequence":180,"environment":181},{"dataset":94,"sequence":183,"environment":184},{"dataset":94,"sequence":186,"environment":184},{"dataset":107,"sequence":188,"environment":189},[363,366,367,368,369],{"name":364,"methodId":365,"linkable":194,"proposed":75,"self":75},"LeGO-LOAM","legoloam2018",{"name":192,"methodId":193,"linkable":194,"proposed":75,"self":75},{"name":201,"methodId":84,"linkable":75,"proposed":75,"self":75},{"name":199,"methodId":84,"linkable":75,"proposed":75,"self":75},{"name":7,"methodId":5,"linkable":194,"proposed":194,"self":194},[371,373,375,377,379,381,382,383,384,386,388,390,392,394,396,398,400,402,404,406],[205,205,205,372,207,205,207,207,205],0.052,[205,205,209,374,207,205,207,207,205],0.129,[205,205,212,376,207,205,207,207,205],0.161,[205,205,158,378,207,205,207,207,205],0.126,[209,205,205,380,207,205,207,207,205],0.248,[209,205,209,84,205,205,207,207,205],[209,205,212,84,205,205,207,207,205],[209,205,158,84,209,205,207,207,205],[212,205,205,385,207,205,207,207,205],0.047,[212,205,209,387,207,205,207,207,205],0.952,[212,205,212,389,207,205,207,207,205],0.928,[212,205,158,391,207,205,207,207,205],0.931,[158,205,205,393,207,205,207,207,205],0.034,[158,205,209,395,207,205,207,207,205],0.071,[158,205,212,397,207,205,207,207,205],0.073,[158,205,158,399,207,205,207,207,205],0.306,[217,205,205,401,207,205,207,207,205],0.029,[217,205,209,403,207,205,207,207,205],0.064,[217,205,212,405,207,205,207,207,205],0.056,[217,205,158,284,207,205,207,207,205],[342,408],"not_run",[352],[],[],[413],"RMS APE (m), median of 5 runs per method; trajectories aligned following Zhang and Scaramuzza [30] with evo (alignment type not stated); x = failed; '-' = not run (NICE-SLAM needs RGB-D and Newer College has monochrome fisheye cameras). NICE-SLAM ran offline on RGB-D simulated from stereo with RAFT, 350 iterations per KeyFrame; it succeeded on MCR in 4 of 5 runs.",{"slug":415,"group":416,"sourceId":5,"sourceLabel":6,"table":417,"selfRows":217,"metrics":418,"seqs":430,"entrants":433,"cells":435,"outcomes":444,"locators":445,"hardware":446,"wordings":448,"notes":449},"loner2023-text-sec-iv-e","loner2023:Text Sec.IV-E","Text Sec.IV-E",[419,422,425,427],{"label":420,"unit":421,"statistic":169,"alignment":45},"average time per tracking step","ms",{"label":423,"unit":424,"statistic":169,"alignment":45},"average time for 50 mapping iterations per KeyFrame (run in parallel with tracking)","s",{"label":426,"unit":421,"statistic":169,"alignment":45},"time per mapping iteration (approximate)",{"label":428,"unit":429,"statistic":170,"alignment":45},"map update rate (approximate)","Hz",[431],{"dataset":170,"sequence":432,"environment":170},"not_reported (Sec. IV-E does not name the sequence used for timing)",[434],{"name":7,"methodId":5,"linkable":194,"proposed":194,"self":194},[436,438,440,442],[205,205,205,437,207,205,205,207,205],14,[205,209,205,439,207,205,205,207,205],2.79,[205,212,205,441,207,205,205,207,205],56,[205,158,205,443,207,205,205,207,205],18,[],[86],[447],"AMD Ryzen 5950X CPU + NVIDIA A6000 GPU",[],[450],"Runtime of LONER in real-time configuration (scans decimated to 5 Hz, one KeyFrame every 3 s, 50 iterations per KeyFrame).",[],1790510658703]