[{"data":1,"prerenderedAt":286},["ShallowReactive",2],{"method-voxgraph2020":3},{"method":4,"reference":64,"equipment":89,"figures":136,"results":137},{"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":26,"sensors":31,"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},"voxgraph2020","Reijgwart et al., 2020","Voxgraph","Voxgraph: Globally Consistent, Volumetric Mapping Using Signed Distance Function Submaps",2020,"recent","C08","full_slam_with_global_correction","Voxgraph 以一組相互重疊的符號距離函數（SDF）子地圖表示環境。前端依固定時間間隔把連續點雲以 voxblox 光線投射整合成 TSDF 子地圖，子地圖完成後再計算歐氏符號距離場（ESDF），並以 marching cubes 取出等值面點。後端以位姿圖最佳化各子地圖的位置與偏航角，約束包括里程計、外部迴圈閉合（例如 DBoW2）以及作者提出的免對應配準約束：把一個子地圖的等值面點轉入相鄰子地圖的 ESDF，直接讀取距離值作為殘差，並依體素權重隨機只取 5% 殘差以降低計算量。系統以地圖為中心，不重新估計完整軌跡，並在搭載 Intel i7-8650U 的六旋翼無人機上以 Ouster OS1 光達或 RealSense D415 即時運作。","Map-centric SDF-submap SLAM: overlapping TSDF\u002FESDF submaps are aligned by a pose graph over x, y, z and yaw using odometry, external loop closures and correspondence-free ESDF registration residuals (iso-surface points of one submap looked up in its neighbour's ESDF), with weight-proportional residual sub-sampling that keeps global optimization real time on an MAV CPU.","full_text_reviewed","peer_reviewed_published","main_body","論文未在施工現場測試。室外實驗在瑞士 Wangen an der Aare 的搜救訓練場，場景含倒塌結構瓦礫堆與穿越建築物的室內外轉換（Fig. 1、Sec. VIII-B1）；室內 RGB-D 實驗在 ETH Zürich 的地下工業空間。場地幾何精度只以 RTK-GNSS 軌跡誤差作為代理指標。其 OS1-64 無人機資料集後來被 molalo2025 與 millane2024nvblox 用作評估資料。",[20,21],"simulation","independent_reference",[23,24,25],"Globally consistent volumetric map computed on the MAV CPU; ATE RMSE 0.83, 0.59, 0.94 and 0.52 m on four 400 m flights, lower than its ROVIO input, VINS-Mono and LOAM (Table I, Sec. VIII-B1)","In simulation, sub-sampling registration residuals down to about 5% showed no observable change in ESDF or trajectory RMSE while solver time fell linearly (Sec. VIII-A, Fig. 4)","External loop closures removed gross map distortion in the RGB-D industrial dataset where submap registration alone could not (Sec. VIII-B2, Fig. 9)",[27,28,29,30],"Higher CPU load than the compared trajectory estimators (230 to 305% versus 125 to 159%), which the authors attribute to computing the volumetric map (Table I, Sec. VIII-B1)","Wide-baseline loops cannot be corrected by submap registration alone and require an external place-recognition source (Sec. VIII-B2)","Field evaluation uses RTK-GNSS trajectory error as a proxy for map quality; reconstruction error against reference geometry is reported only in simulation (Sec. VIII-A, VIII-B1)","Submap poses are optimized in 4 DoF, relying on gravity-aligned odometry (Sec. III)",[32,33,34,35],"3D LiDAR Ouster OS1 (64-beam) in the outdoor MAV field experiment","RGB-D camera Intel RealSense D415 (pointclouds) in the indoor MAV experiment","Odometry input from a time-synchronized camera-IMU running ROVIO visual-inertial odometry; VI-sensor stereo and IMU in the indoor dataset","RTK-GNSS used only as trajectory ground truth",[37,38,39],"UAV (hexacopter MAV; four outdoor flights of about 400 m at a search and rescue training site)","UAV (MAV with RGB-D camera and VI-sensor in an underground industrial space at ETH Zurich; dataset from C-blox)","simulation (RotorS, LiDAR-equipped MAV flying around a multi-story building)","Pose-graph nonlinear least squares over submap poses in R3 x SO(2) (x, y, z, yaw; roll and pitch taken from gravity-aligned odometry) with odometry and loop-closure terms (Mahalanobis) and registration terms (weighted squared ESDF distance); per-scan poses come from the external visual-inertial odometry (ROVIO) and are kept fixed relative to their submap","Correspondence-free submap-to-submap registration: iso-surface points of one submap are transformed into the overlapping submap and its trilinearly interpolated ESDF value is the residual; overlapping pairs found with axis-aligned bounding boxes; residuals randomly sub-sampled per solver iteration with probability proportional to voxel weight (5% sampling ratio used)","discrete poses; per-scan odometry poses stored relative to their submap","LiDAR undistortion runs as a component in the field experiment (listed in the CPU breakdown, Fig. 6); the method is not described","External, source-agnostic loop closures between two sensor frames are converted into constraints between the submaps that contain them (DBoW2 place recognition on VI-sensor images in the RGB-D experiment); registration constraints link overlapping submaps, while wide loops rely on the external loop closures (Secs. VI-B, VIII-B2)","Pose graph over all submap poses, re-solved when each new submap is added; maximum global optimization time about 4 s in the field experiments","collection of overlapping TSDF submaps built at a fixed frequency by ray casting into spatially hashed voxel blocks (voxblox), each with an ESDF and marching-cubes iso-surface points; submaps can be fused into one global map","none (assumes gravity direction observable from the odometry front end)","globally aligned SDF submap collection and fused global volumetric map; trajectory obtained by projecting submap-relative odometry with the optimised submap poses","CPU only; Intel NUC Core i7-8650U carried by the MAV; fewer than 3 CPU threads (230 to 305% where 100% is one hyper-thread); global optimization uses 44% of one core; RGB-D dataset processed on a desktop CPU with 5 cm voxels (Sec. VIII-B)","https:\u002F\u002Fgithub.com\u002Fethz-asl\u002Fvoxgraph","BSD-2-Clause (LICENSE file checked)",[53,57,60],{"relation":54,"title":55,"doi_or_url":56},"preprint","Voxgraph (arXiv v1, RA-L accepted version header)","https:\u002F\u002Farxiv.org\u002Fabs\u002F2004.13154",{"relation":58,"title":59,"doi_or_url":50},"code_release","ethz-asl\u002Fvoxgraph",{"relation":61,"title":62,"doi_or_url":63},"predecessor_method","C-blox: A Scalable and Consistent TSDF-based Dense Mapping Approach (IROS 2018), cited as the authors' previous submap work [26]","10.1109\u002FIROS.2018.8593427",{"id":5,"kind":65,"shortName":7,"title":8,"authors":66,"year":9,"venue":73,"venueType":74,"publisher":75,"volumeIssuePages":76,"doi":77,"arxivId":78,"url":79,"firstPublicDate":80,"publicationStatus":16,"metadataStatus":81,"fulltextStatus":15,"era":10,"classicReason":82,"codeUrl":50,"cluster":11,"topics":83,"mdpi":84,"verification":85,"label":6,"fulltextRoute":86,"versionRead":87,"addedByCensus":88},"method",[67,68,69,70,71,72],"Victor Reijgwart","Alexander Millane","Helen Oleynikova","Roland Siegwart","Cesar Cadena","Juan Nieto","IEEE Robotics and Automation Letters","journal","IEEE","5(1):227-234","10.1109\u002Flra.2019.2953859","2004.13154","https:\u002F\u002Fapi.crossref.org\u002Fworks\u002F10.1109\u002FLRA.2019.2953859","2019-11-25","metadata_verified","not_applicable",[11],false,"corrected","arXiv","arXiv 2004.13154v1 (2020-04-27), header 'IEEE Robotics and Automation Letters. Preprint version. Accepted October, 2019'; IEEE version of record not compared",true,[90,97,104,108,112,117,123,127,133],{"category":91,"model":92,"canonical":92,"role":93,"dataset":94,"specs":95,"locator":96},"lidar","Ouster OS1","method input",null,"64 beam (Fig. 5 caption)","Sec. VIII-B; Fig. 5",{"category":98,"model":99,"canonical":100,"role":93,"dataset":101,"specs":102,"locator":103},"rgbd","Intel RealSense D415 Depth Camera","Intel RealSense D415","indoor MAV industrial dataset of C-blox [26]","produces RGB-D data; indoor dataset processed with 5 cm voxels","Sec. VIII-B, VIII-B2",{"category":105,"model":106,"canonical":106,"role":93,"dataset":94,"specs":107,"locator":96},"camera","monocular camera (model not reported)","time-synchronized with IMU; feeds ROVIO visual-inertial odometry",{"category":109,"model":110,"canonical":110,"role":93,"dataset":94,"specs":111,"locator":96},"imu","time-synchronized IMU (model not reported)","feeds ROVIO visual-inertial odometry",{"category":113,"model":114,"canonical":114,"role":93,"dataset":101,"specs":115,"locator":116},"stereo_camera","VI-sensor (visual-inertial sensor of Nikolic et al. [36])","images used for DBoW2 loop closures and odometry in the indoor dataset","Sec. VIII-B2",{"category":118,"model":119,"canonical":119,"role":120,"dataset":94,"specs":121,"locator":122},"gnss","RTK-GNSS system (model not reported)","reference or ground truth","attached to the MAV; used for trajectory evaluation only","Sec. VIII-B1; Fig. 5",{"category":124,"model":125,"canonical":125,"role":93,"dataset":94,"specs":126,"locator":96},"platform","hexacopter MAV","carries 64-beam LiDAR, monocular camera, synchronized IMU and RTK-GNSS",{"category":128,"model":129,"canonical":129,"role":130,"dataset":94,"specs":131,"locator":132},"compute","Intel NUC Core i7-8650U","compute for runtime","carried by the MAV; all calculations on board; 800% maximum load (8 hyper-threads)","Abstract; Sec. VIII-B",{"category":128,"model":134,"canonical":134,"role":130,"dataset":94,"specs":135,"locator":116},"desktop CPU (model not reported)","used to process the RGB-D industrial dataset",[],{"totalRows":138,"groupCount":139,"groups":140,"others":285},10,2,[141,250],{"slug":142,"group":143,"sourceId":5,"sourceLabel":6,"table":144,"selfRows":145,"metrics":146,"seqs":157,"entrants":168,"cells":178,"outcomes":241,"locators":243,"hardware":244,"wordings":246,"notes":247},"voxgraph2020-table-i","voxgraph2020:Table I","Table I",8,[147,152],{"label":148,"unit":149,"statistic":150,"alignment":151},"RMSE (m)","m","RMSE","SE3",{"label":153,"unit":154,"statistic":155,"alignment":156},"CPU (%)","%","mean","not_reported",[158,162,164,166],{"dataset":159,"sequence":160,"environment":161},"Voxgraph MAV field dataset (this paper)","t0","outdoor search and rescue training site with rubble and indoor-outdoor transitions, hexacopter MAV",{"dataset":159,"sequence":163,"environment":161},"t1",{"dataset":159,"sequence":165,"environment":161},"t2",{"dataset":159,"sequence":167,"environment":161},"t3",[169,171,172,175],{"name":170,"methodId":94,"linkable":84,"proposed":84,"self":84},"ROVIO (odometry input to Voxgraph)",{"name":7,"methodId":5,"linkable":88,"proposed":88,"self":88},{"name":173,"methodId":174,"linkable":88,"proposed":84,"self":84},"Vins-Mono","vinsmono2018",{"name":176,"methodId":177,"linkable":88,"proposed":84,"self":84},"Loam","loam2017_auro",[179,183,186,188,191,193,195,196,198,200,202,204,206,208,210,211,213,215,217,219,221,222,224,225,227,229,231,233,235,236,238,239],[180,180,180,181,182,180,182,182,180],0,4.55,-1,[184,180,180,185,182,180,182,182,180],1,0.83,[139,180,180,187,182,180,182,182,180],5.51,[189,180,180,190,182,180,182,182,180],3,2.64,[180,180,184,192,182,180,182,182,180],1.3,[184,180,184,194,182,180,182,182,180],0.59,[139,180,184,94,180,180,182,182,180],[189,180,184,197,182,180,182,182,180],1.29,[180,180,139,199,182,180,182,182,180],3.52,[184,180,139,201,182,180,182,182,180],0.94,[139,180,139,203,182,180,182,182,180],1.11,[189,180,139,205,182,180,182,182,180],6.43,[180,180,189,207,182,180,182,182,180],2.25,[184,180,189,209,182,180,182,182,180],0.52,[139,180,189,94,180,180,182,182,180],[189,180,189,212,182,180,182,182,180],3.55,[180,184,180,214,182,180,180,182,184],47,[184,184,180,216,182,180,180,182,184],265,[139,184,180,218,182,180,180,182,184],159,[189,184,180,220,182,180,180,182,184],131,[180,184,184,214,182,180,180,182,184],[184,184,184,223,182,180,180,182,184],230,[139,184,184,94,180,180,180,182,184],[189,184,184,226,182,180,180,182,184],129,[180,184,139,228,182,180,180,182,184],48,[184,184,139,230,182,180,180,182,184],305,[139,184,139,232,182,180,180,182,184],157,[189,184,139,234,182,180,180,182,184],125,[180,184,189,214,182,180,180,182,184],[184,184,189,237,182,180,180,182,184],286,[139,184,189,94,180,180,180,182,184],[189,184,189,240,182,180,180,182,184],127,[242],"diverged",[144],[245],"Intel NUC Core i7-8650U (on board)",[],[248,249],"Four MAV flights (about 400 m each) at the Wangen an der Aare search and rescue training site; RTK-GNSS ground truth; each system run 10 times and averages reported; 4-DoF alignment for visual-inertial systems and 6-DoF for LOAM; '-' means the estimator diverged","Four MAV flights (about 400 m each) at the Wangen an der Aare search and rescue training site; RTK-GNSS ground truth; each system run 10 times and averages reported; 4-DoF alignment for visual-inertial systems and 6-DoF for LOAM; '-' means the estimator diverged; CPU where 100% is one fully used hyper-thread",{"slug":251,"group":252,"sourceId":5,"sourceLabel":6,"table":253,"selfRows":139,"metrics":254,"seqs":261,"entrants":266,"cells":270,"outcomes":275,"locators":277,"hardware":280,"wordings":281,"notes":282},"voxgraph2020-text-sec-viii-b1","voxgraph2020:Text Sec.VIII-B1","Text Sec.VIII-B1",[255,259],{"label":256,"unit":257,"statistic":258,"alignment":156},"maximum runtime for global optimization","s","max",{"label":260,"unit":154,"statistic":156,"alignment":156},"global optimizations use 44% of a single CPU core",[262,264],{"dataset":159,"sequence":263,"environment":161},"t0 to t3",{"dataset":159,"sequence":265,"environment":161},"typical flight",[267,268],{"name":7,"methodId":5,"linkable":88,"proposed":88,"self":88},{"name":269,"methodId":5,"linkable":88,"proposed":88,"self":88},"Voxgraph (global optimization only)",[271,273],[180,180,180,272,180,180,180,182,180],4,[184,184,184,274,182,184,180,182,184],44,[276],"other: approximate value ('~4 s' in Sec. VIII-B1; the abstract says less than 4 s)",[278,279],"Sec. VIII-B1; Fig. 8; Abstract (120x80 m map optimized in less than 4 s)","Sec. VIII-B1; Fig. 6",[245],[],[283,284],"Global optimization time over 10 trials of each of 4 trajectories; text gives only the maximum","Share of CPU used by the global optimization during a typical flight",[],1790510662488]