[{"data":1,"prerenderedAt":434},["ShallowReactive",2],{"method-millane2024nvblox":3},{"method":4,"reference":56,"equipment":79,"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":23,"limitations":28,"sensors":33,"platform":36,"estimator":40,"association":41,"timeModel":42,"deskew":43,"loopClosure":42,"globalOptimization":44,"mapRepresentation":45,"prior":46,"outputGeometry":47,"compute":48,"codeUrl":49,"codeLicense":50,"relatedVersions":51},"millane2024nvblox","Millane et al., 2024","nvblox","nvblox: GPU-Accelerated Incremental Signed Distance Field Mapping",2024,"recent","C12","map_representation_or_reconstruction","nvblox 將 Voxblox 的分層體素地圖移到 GPU：以雜湊表索引 8x8x8 體素區塊，並行更新 TSDF 或佔據層，定期以平行 marching cubes 產生網格；另提出以區塊內掃掠與跨區塊傳遞交替進行的增量式 GPU ESDF 演算法，採完整歐氏距離而非近似距離。支援 RGB-D 與 LiDAR，並可在嵌入式 GPU 上執行。","A GPU library that ports block-hashed TSDF\u002Foccupancy\u002FESDF\u002Fmesh mapping to the GPU, with an incremental parallel full-Euclidean ESDF, supporting RGB-D and LiDAR.","full_text_reviewed","peer_reviewed_published","main_body","示範場景為辦公室建築中的地面機器人、無人機 LiDAR 資料與機械手臂，未在營建場域驗證；公分級 TSDF 可即時運算的結果對現場快速建網格有參考價值，但未證明量測精度（推論）。",[20,21,22],"public_benchmark","simulation","controlled_experiment",[24,25,26,27],"Up to 177x faster TSDF integration and on average 31x faster incremental ESDF than voxblox (Tables I-II); the TSDF figure compares nvblox surface-only integration with voxblox surface plus color integration (Table I caption).","Even at 1 cm TSDF and 2 cm ESDF resolution, nvblox was faster than voxblox at 10 cm on the Replica office0 sequence (Sec. V, Fig. 6).","Average median ESDF error 0.06 m versus 0.07 m (voxblox) and 0.06 m (Fiesta) while being 7x faster than Fiesta (Table II).","GPU distance queries reach 6.2 to 7.3 billion per second (correlated) on the desktop and 0.3 to 0.8 billion on the Jetson (Table III).",[29,30,31,32],"Evaluation focuses on runtime and ESDF accuracy for planning; surface accuracy of the mesh against an independent reference is not reported (reviewer reading of Sec. V).","(inference) Requires an NVIDIA GPU.","On the Cow and Lady dataset nvblox had a higher median ESDF error (0.09 m) than voxblox (0.06 m) and Fiesta (0.07 m) (Table II).","(reviewer check) The drone LiDAR example is stated at 5 cm resolution in Sec. V-E but 10 cm in the Fig. 7 caption, also in the ICRA version.",[34,35],"RGB-D","3D LiDAR",[37,38,39],"handheld (Intel RealSense D455, Fig. 1)","UAV (drone LiDAR dataset from Voxgraph [6], Fig. 7)","ground robot in an office building (locomotion type not stated, Fig. 8)","not_applicable (poses supplied; e.g., FAST-LIO for the drone LiDAR example)","projective TSDF update of voxels in view (LiDAR depth images with linear interpolation)","not_applicable","not_reported","none","block-hashed voxel layers (8x8x8 voxels per block) on GPU: TSDF, ESDF, occupancy, color, mesh","external poses","TSDF, ESDF, occupancy and marching-cubes mesh","GPU: desktop (Intel i9, NVIDIA RTX 3090 Ti), laptop (Intel i7, RTX 3000 Mobile) and Jetson AGX Xavier; TSDF integration 0.4, 0.6 and 1.6 ms per frame on Replica at 5 cm (Table I); drone LiDAR integration under 7 ms per scan on the laptop and under 20 ms on the Jetson (Sec. V-E, Fig. 7)","https:\u002F\u002Fgithub.com\u002Fnvidia-isaac\u002Fnvblox","Apache-2.0 (repository LICENSE.md)",[52],{"relation":53,"title":54,"doi_or_url":55},"preprint","arXiv:2311.00626 (v1 2023-11-01, v2)","https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.00626",{"id":5,"kind":57,"shortName":7,"title":8,"authors":58,"year":9,"venue":66,"venueType":67,"publisher":68,"volumeIssuePages":69,"doi":70,"arxivId":71,"url":55,"firstPublicDate":72,"publicationStatus":16,"metadataStatus":73,"fulltextStatus":15,"era":10,"classicReason":42,"codeUrl":49,"cluster":11,"topics":74,"mdpi":75,"verification":76,"label":6,"fulltextRoute":77,"versionRead":78,"addedByCensus":75},"software",[59,60,61,62,63,64,65],"Alexander Millane","Helen Oleynikova","Emilie Wirbel","Remo Steiner","Vikram Ramasamy","David Tingdahl","Roland Siegwart","2024 IEEE International Conference on Robotics and Automation (ICRA)","conference","IEEE","pp. 2698-2705","10.1109\u002Ficra57147.2024.10611532","2311.00626","2023-11-01","metadata_verified",[11],false,"confirmed","arXiv","arXiv v2 (2024-03-15, 'Accepted to ICRA 2024') read in full; ICRA 2024 version of record pp. 2698-2705 obtained via NTU access and Tables I-II and Sec. V-E checked (identical values)",[80,87,94,100,102],{"category":81,"model":82,"canonical":82,"role":83,"dataset":84,"specs":85,"locator":86},"rgbd","Intel RealSense Depth Camera D455","method input",null,"handheld; real-time reconstruction on an embedded GPU","Fig. 1",{"category":88,"model":89,"canonical":89,"role":90,"dataset":91,"specs":92,"locator":93},"lidar","Ouster OS1 (64-beam)","dataset sensor","drone LiDAR dataset of Voxgraph [6]","64-beam; integrated up to 25 m","Sec. V-E, Fig. 7",{"category":95,"model":96,"canonical":96,"role":97,"dataset":84,"specs":98,"locator":99},"compute","Intel i9 CPU + NVIDIA RTX 3090 Ti (Desktop)","compute for runtime","not_reported beyond model","Sec. V",{"category":95,"model":101,"canonical":101,"role":97,"dataset":84,"specs":98,"locator":99},"Intel i7 CPU + NVIDIA RTX 3000 Mobile (Laptop)",{"category":95,"model":103,"canonical":103,"role":97,"dataset":84,"specs":104,"locator":99},"NVIDIA Jetson AGX Xavier","embedded GPU platform",[],{"totalRows":107,"groupCount":108,"groups":109,"others":433},58,4,[110,258,368,408],{"slug":111,"group":112,"sourceId":5,"sourceLabel":6,"table":113,"selfRows":114,"metrics":115,"seqs":128,"entrants":136,"cells":142,"outcomes":249,"locators":251,"hardware":252,"wordings":255,"notes":256},"millane2024nvblox-table-i","millane2024nvblox:Table I","Table I",30,[116,120,122,124,126],{"label":117,"unit":118,"statistic":119,"alignment":42},"ESDF component runtime (ms), computed every 4 frames","ms","mean",{"label":121,"unit":118,"statistic":119,"alignment":42},"TSDF component runtime (ms)",{"label":123,"unit":118,"statistic":119,"alignment":42},"Color component runtime (ms)",{"label":125,"unit":118,"statistic":119,"alignment":42},"TSDF+Color component runtime (ms)",{"label":127,"unit":118,"statistic":119,"alignment":42},"Mesh component runtime (ms), computed every 4 frames",[129,133],{"dataset":130,"sequence":131,"environment":132},"Replica","average over sequences","synthetic rooms (photorealistic renderings)",{"dataset":134,"sequence":131,"environment":135},"Redwood","real consumer depth camera scans",[137,139],{"name":7,"methodId":5,"linkable":138,"proposed":138,"self":138},true,{"name":140,"methodId":141,"linkable":138,"proposed":75,"self":75},"voxblox","oleynikova2017voxblox",[143,147,150,152,154,157,159,161,162,164,165,167,168,170,171,173,174,176,177,180,182,184,186,188,190,191,193,194,196,198,200,202,204,206,208,209,211,213,214,215,216,218,219,221,222,223,224,226,227,229,231,233,235,237,239,240,242,243,245,247],[144,144,144,145,146,144,144,146,144],0,1.9,-1,[148,144,144,149,146,144,144,146,144],1,163.2,[144,144,144,151,146,144,148,146,144],3.6,[148,144,144,153,146,144,148,146,144],291.5,[144,144,144,155,146,144,156,146,144],8.4,2,[148,144,144,158,146,144,156,146,144],231.6,[144,148,144,160,146,144,144,146,144],0.4,[148,148,144,84,144,144,144,146,144],[144,148,144,163,146,144,148,146,144],0.6,[148,148,144,84,144,144,148,146,144],[144,148,144,166,146,144,156,146,144],1.6,[148,148,144,84,144,144,156,146,144],[144,156,144,169,146,144,144,146,144],1.7,[148,156,144,84,144,144,144,146,144],[144,156,144,172,146,144,148,146,144],2.5,[148,156,144,84,144,144,148,146,144],[144,156,144,175,146,144,156,146,144],4.2,[148,156,144,84,144,144,156,146,144],[144,178,144,179,146,144,144,146,144],3,2.1,[148,178,144,181,146,144,144,146,144],86.7,[144,178,144,183,146,144,148,146,144],3.2,[148,178,144,185,146,144,148,146,144],106.6,[144,178,144,187,146,144,156,146,144],5.8,[148,178,144,189,146,144,156,146,144],226.7,[144,108,144,166,146,144,144,146,144],[148,108,144,192,146,144,144,146,144],6.2,[144,108,144,108,146,144,148,146,144],[148,108,144,195,146,144,148,146,144],12,[144,108,144,197,146,144,156,146,144],12.3,[148,108,144,199,146,144,156,146,144],15.4,[144,144,148,201,146,144,144,146,144],1.5,[148,144,148,203,146,144,144,146,144],29.1,[144,144,148,205,146,144,148,146,144],2.6,[148,144,148,207,146,144,148,146,144],46.5,[144,144,148,175,146,144,156,146,144],[148,144,148,210,146,144,156,146,144],38.7,[144,148,148,212,146,144,144,146,144],0.2,[148,148,148,84,144,144,144,146,144],[144,148,148,212,146,144,148,146,144],[148,148,148,84,144,144,148,146,144],[144,148,148,217,146,144,156,146,144],0.5,[148,148,148,84,144,144,156,146,144],[144,156,148,220,146,144,144,146,144],1.1,[148,156,148,84,144,144,144,146,144],[144,156,148,166,146,144,148,146,144],[148,156,148,84,144,144,148,146,144],[144,156,148,225,146,144,156,146,144],2.4,[148,156,148,84,144,144,156,146,144],[144,178,148,228,146,144,144,146,144],1.3,[148,178,148,230,146,144,144,146,144],38.4,[144,178,148,232,146,144,148,146,144],1.8,[148,178,148,234,146,144,148,146,144],33.6,[144,178,148,236,146,144,156,146,144],2.9,[148,178,148,238,146,144,156,146,144],76.7,[144,108,148,163,146,144,144,146,144],[148,108,148,241,146,144,144,146,144],12.7,[144,108,148,201,146,144,148,146,144],[148,108,148,244,146,144,148,146,144],15.8,[144,108,148,246,146,144,156,146,144],2.7,[148,108,148,248,146,144,156,146,144],23,[250],"not_reported (voxblox does not separate TSDF and color integration)",[113],[253,254,103],"Desktop: Intel i9 CPU, NVIDIA RTX 3090 Ti","Laptop: Intel i7 CPU, RTX 3000 Mobile GPU",[],[257],"Component runtimes averaged over 8 Replica and 5 Redwood sequences at 5 cm voxels; ESDF and mesh computed every 4 frames; voxblox does not separate TSDF and color integration, so its TSDF and Color cells are empty; speed-up column not extracted",{"slug":259,"group":260,"sourceId":5,"sourceLabel":6,"table":261,"selfRows":262,"metrics":263,"seqs":273,"entrants":291,"cells":296,"outcomes":362,"locators":363,"hardware":364,"wordings":365,"notes":366},"millane2024nvblox-table-ii","millane2024nvblox:Table II","Table II",14,[264,268,270,272],{"label":265,"unit":266,"statistic":267,"alignment":42},"Median ESDF error (m)","m","median",{"label":269,"unit":118,"statistic":43,"alignment":42},"ESDF runtime (ms)",{"label":271,"unit":266,"statistic":119,"alignment":42},"Median ESDF error (m), Average row (arithmetic mean of the six per-sequence medians)",{"label":269,"unit":118,"statistic":119,"alignment":42},[274,277,279,281,283,285,288],{"dataset":134,"sequence":275,"environment":276},"apartment","indoor RGB-D scans",{"dataset":134,"sequence":278,"environment":276},"bedroom",{"dataset":134,"sequence":280,"environment":276},"boardroom",{"dataset":134,"sequence":282,"environment":276},"lobby",{"dataset":134,"sequence":284,"environment":276},"loft",{"dataset":286,"sequence":287,"environment":276},"Cow and lady","-",{"dataset":289,"sequence":290,"environment":276},"Average","all",[292,293,294],{"name":7,"methodId":5,"linkable":138,"proposed":138,"self":138},{"name":140,"methodId":141,"linkable":138,"proposed":75,"self":75},{"name":295,"methodId":84,"linkable":75,"proposed":75,"self":75},"Fiesta",[297,299,300,302,304,306,308,310,312,313,315,317,319,320,321,323,324,325,326,328,329,330,332,333,335,336,337,338,340,341,342,345,347,348,350,352,354,356,357,358,359,360],[144,144,144,298,146,144,146,146,144],0.04,[144,148,144,169,146,144,144,146,144],[148,144,144,301,146,144,146,146,144],0.06,[148,148,144,303,146,144,144,146,144],25,[156,144,144,305,146,144,146,146,144],0.05,[156,148,144,307,146,144,144,146,144],5.5,[144,144,148,309,146,144,146,146,144],0.02,[144,148,148,311,146,144,144,146,144],1.4,[148,144,148,305,146,144,146,146,144],[148,148,148,314,146,144,144,146,144],22,[156,144,148,316,146,144,146,146,144],0.03,[156,148,148,318,146,144,144,146,144],3.4,[144,144,156,301,146,144,146,146,144],[144,148,156,169,146,144,144,146,144],[148,144,156,322,146,144,146,146,144],0.08,[148,148,156,114,146,144,144,146,144],[156,144,156,301,146,144,146,146,144],[156,148,156,108,146,144,144,146,144],[144,144,178,327,146,144,146,146,144],0.1,[144,148,178,179,146,144,144,146,144],[148,144,178,327,146,144,146,146,144],[148,148,178,331,146,144,144,146,144],34,[156,144,178,322,146,144,146,146,144],[156,148,178,334,146,144,144,146,144],5.2,[144,144,108,298,146,144,146,146,144],[144,148,108,232,146,144,144,146,144],[148,144,108,322,146,144,146,146,144],[148,148,108,339,146,144,144,146,144],48,[156,144,108,298,146,144,146,146,144],[156,148,108,155,146,144,144,146,144],[144,144,343,344,146,144,146,146,144],5,0.09,[144,148,343,346,146,144,144,146,144],2.8,[148,144,343,301,146,144,146,146,144],[148,148,343,349,146,144,144,146,144],190,[156,144,343,351,146,144,146,146,144],0.07,[156,148,343,353,146,144,144,146,144],52,[144,156,355,301,146,144,146,146,144],6,[144,178,355,145,146,144,144,146,144],[148,156,355,351,146,144,146,146,144],[148,178,355,107,146,144,144,146,144],[156,156,355,301,146,144,146,146,144],[156,178,355,361,146,144,144,146,144],13,[],[261],[253],[],[367],"Incremental ESDF on the Desktop platform; error is the median absolute voxel-wise difference to a voxelized ESDF ground truth computed from reconstructed voxel centres to the dataset ground-truth surface; the table cites Redwood as [3] although the caption cites [32]",{"slug":369,"group":370,"sourceId":5,"sourceLabel":6,"table":371,"selfRows":195,"metrics":372,"seqs":378,"entrants":382,"cells":384,"outcomes":402,"locators":403,"hardware":404,"wordings":405,"notes":406},"millane2024nvblox-table-iii","millane2024nvblox:Table III","Table III",[373,376],{"label":374,"unit":375,"statistic":119,"alignment":42},"distance queries per second (10^9), cor.","10^9 queries\u002Fs",{"label":377,"unit":375,"statistic":119,"alignment":42},"distance queries per second (10^9), uncor.",[379,380],{"dataset":134,"sequence":131,"environment":276},{"dataset":381,"sequence":131,"environment":276},"Sun3D",[383],{"name":7,"methodId":5,"linkable":138,"proposed":138,"self":138},[385,386,388,389,390,392,393,395,396,397,398,400],[144,144,144,192,146,144,144,146,144],[144,148,144,387,146,144,144,146,144],3.3,[144,144,144,169,146,144,148,146,144],[144,148,144,228,146,144,148,146,144],[144,144,144,391,146,144,156,146,144],0.8,[144,148,144,217,146,144,156,146,144],[144,144,148,394,146,144,144,146,144],7.3,[144,148,148,387,146,144,144,146,144],[144,144,148,232,146,144,148,146,144],[144,148,148,220,146,144,148,146,144],[144,144,148,399,146,144,156,146,144],0.7,[144,148,148,401,146,144,156,146,144],0.3,[],[371],[253,254,103],[],[407],"Distance query throughput on GPU averaged over several sequences; cor. = spatially correlated query points, uncor. = uncorrelated",{"slug":409,"group":410,"sourceId":5,"sourceLabel":6,"table":411,"selfRows":156,"metrics":412,"seqs":416,"entrants":419,"cells":421,"outcomes":426,"locators":427,"hardware":429,"wordings":430,"notes":431},"millane2024nvblox-text-sec-v-e","millane2024nvblox:Text Sec. V-E","Text Sec. V-E",[413],{"label":414,"unit":118,"statistic":415,"alignment":42},"LiDAR integration time per scan (less than)","max",[417],{"dataset":91,"sequence":43,"environment":418},"large-scale outdoor, flying robot",[420],{"name":7,"methodId":5,"linkable":138,"proposed":138,"self":138},[422,424],[144,144,144,423,146,144,144,146,144],7,[144,144,144,425,146,148,148,146,144],20,[],[93,428],"Fig. 7 caption",[254,103],[],[432],"Drone dataset with a 64-beam Ouster OS1, FAST-LIO poses, 25 m integration range; resolution given as 5 cm in the text but 10 cm in the Fig. 7 caption",[],1790510659200]