[{"data":1,"prerenderedAt":231},["ShallowReactive",2],{"method-nubert2022learninglocalizability":3},{"method":4,"reference":50,"equipment":70,"figures":109,"results":110},{"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":35,"estimator":37,"association":38,"timeModel":39,"deskew":40,"loopClosure":39,"globalOptimization":41,"mapRepresentation":39,"prior":42,"outputGeometry":39,"compute":43,"codeUrl":44,"codeLicense":45,"relatedVersions":46},"nubert2022learninglocalizability","Nubert et al., 2022b","Learning-based localizability","Learning-based Localizability Estimation for Robust LiDAR Localization",2022,"recent","C13","registration_component","本文以神經網路直接由單一 LiDAR 掃描預測掃描對掃描配準在六個自由度上是否可定位，不需先建立對應或求解配準最佳化即可提早偵測失效。網路只用模擬資料訓練，並取代 CompSLAM 中以特徵值門檻判斷退化的模組；在礦坑隧道、開闊混凝土場地與辦公室的現地測試中，以地圖與曲線定性展示同一網路不需重新調參，也能轉用到 Ouster OS0-128。","Neural network trained on simulation predicts scan-to-scan localizability from raw LiDAR data before registration.","full_text_reviewed","peer_reviewed_published","supplementary","現地測試之一在瑞士 Rümlang，原文先稱其為鋪面工作場址（paved work site）上的開闊場地，後又稱為工作場址旁的開闊場地；機器人由棚架下出發，走上周圍沒有幾何特徵的大片混凝土地面，掃描對掃描配準在 x、y 與偏航方向退化。原文未稱其為施工中工地，因此不歸類為 real_construction_site。另兩處為 Seemühle 礦坑隧道，以及 ETH 兼含室內空間與屋頂露台的辦公室環境（arXiv v2 的 Sec. VI-B）。",[20,21,22],"simulation","underground_or_tunnel","completed_building",[24,25,26,27],"No environment-specific threshold tuning; tested on two sensor types without modification (abstract)","On the simulated tunnel test set ResUNet reached F1 0.517 and accuracy 0.854 versus 0.214 and 0.809 for PointNet (Table IV)","In the Seemühle mine tunnel an eigenvalue threshold of at least 110 was needed for the classical detector, while the network flagged y-axis degeneracy without tuning (Fig. 4)","The same network worked with an Ouster OS0-128 although smallest-eigenvalue scales differed markedly from the VLP-16 (Sec. VI-C, Fig. 7)",[29,30,31,32],"Detection only; mitigation relies on the downstream fusion framework (CompSLAM with leg odometry) (Sec. VI-B)","Precision and recall are relatively low for z, roll and pitch because the training set has few non-localizable examples in these directions; test-set precision 0.398 (Sec. VI-A, Table IV)","Binary per-axis labels; predicting a full 6-DOF covariance for robots misaligned with the environment is left to future work (Sec. VII)","Field detection results are shown qualitatively only, with no quantitative metrics on real data (reviewer observation, Sec. VI-B, VI-C)",[34],"3D LiDAR (Velodyne VLP-16; Ouster OS0-128 for sensor-transfer test)",[36,20],"legged (ANYmal C)","sparse 3D convolutional ResUNet feature extractor (MinkowskiEngine; 4000 sampled points, 0.2 m voxels) with global max pooling and a 5-layer MLP giving six sigmoid outputs; multi-label binary classification of localizability along x, y, z, roll, pitch and yaw trained with binary cross-entropy; per-direction probability thresholds chosen from a validation precision-recall curve","not_applicable (prediction before registration)","not_applicable","not_reported","none","trained only on simulated scans from 15 environments (CAD meshes and meshed cave scans) rendered in Gazebo with a simulated VLP-16; labels from Monte Carlo sampling of 200 child scans registered by point-to-plane ICP and thresholded at 0.1 m and 2 deg","inference 28 ms per scan on CPU (Intel i7 10700F) and 13 ms on an Nvidia RTX3070 GPU in a ROS node without code optimization; training about 4 h on an Nvidia RTX3090",null,"not_verified",[47],{"relation":48,"title":8,"doi_or_url":49},"preprint","https:\u002F\u002Farxiv.org\u002Fabs\u002F2203.05698",{"id":5,"kind":51,"shortName":7,"title":8,"authors":52,"year":9,"venue":57,"venueType":58,"publisher":59,"volumeIssuePages":60,"doi":61,"arxivId":62,"url":49,"firstPublicDate":63,"publicationStatus":16,"metadataStatus":64,"fulltextStatus":15,"era":10,"classicReason":39,"codeUrl":44,"cluster":11,"topics":65,"mdpi":66,"verification":67,"label":6,"fulltextRoute":68,"versionRead":69,"addedByCensus":66},"component",[53,54,55,56],"Julian Nubert","Etienne Walther","Shehryar Khattak","Marco Hutter","2022 IEEE\u002FRSJ International Conference on Intelligent Robots and Systems (IROS)","conference","IEEE","pp. 17-24","10.1109\u002Firos47612.2022.9982257","2203.05698","2022-03-11","metadata_verified",[11],false,"corrected","arXiv","arXiv v2 (2022-08-01; v1 2022-03-11 exists); IEEE IROS 2022 version of record not compared",[71,77,81,86,92,96,103,106],{"category":72,"model":73,"canonical":73,"role":74,"dataset":44,"specs":75,"locator":76},"lidar","Velodyne VLP-16","method input","16 beams, 30 deg vertical field of view; its simulated model generated all training scans in Gazebo","Sec. V-A2, VI-B, VI-C",{"category":72,"model":78,"canonical":78,"role":74,"dataset":44,"specs":79,"locator":80},"Ouster OS0-128","128 beams, 90 deg vertical field of view; unseen in training, used for the generalization test","Sec. VI-C",{"category":82,"model":83,"canonical":83,"role":74,"dataset":44,"specs":84,"locator":85},"platform","ANYmal-C","quadrupedal robot; kinematic leg odometry fused in CompSLAM along non-localizable directions","Sec. I, VI-B",{"category":87,"model":88,"canonical":88,"role":89,"dataset":44,"specs":90,"locator":91},"tls_scanner","Leica RTC360","reference or ground truth","used with the BLK2GO to record the ground-truth map of the mine tunnel","Sec. VI-A",{"category":93,"model":94,"canonical":94,"role":89,"dataset":44,"specs":95,"locator":91},"mobile_scanner_device","Leica BLK2GO","used with the RTC360 to record the ground-truth map of the mine tunnel",{"category":97,"model":98,"canonical":99,"role":100,"dataset":44,"specs":101,"locator":102},"compute","Nvidia RTX3090","NVIDIA RTX 3090","compute for runtime","training in about 4 h","Sec. V-B",{"category":97,"model":104,"canonical":104,"role":100,"dataset":44,"specs":105,"locator":102},"Intel i7 10700F","CPU-only inference 28 ms",{"category":97,"model":107,"canonical":107,"role":100,"dataset":44,"specs":108,"locator":102},"Nvidia RTX3070","GPU inference 13 ms",[],{"totalRows":111,"groupCount":112,"groups":113,"others":230},14,2,[114,202],{"slug":115,"group":116,"sourceId":5,"sourceLabel":6,"table":117,"selfRows":118,"metrics":119,"seqs":130,"entrants":139,"cells":145,"outcomes":196,"locators":197,"hardware":198,"wordings":199,"notes":200},"nubert2022learninglocalizability-table-iv","nubert2022learninglocalizability:Table IV","Table IV",12,[120,124,126,128],{"label":121,"unit":122,"statistic":123,"alignment":41},"Accuracy","fraction","mean",{"label":125,"unit":122,"statistic":123,"alignment":41},"F1-score",{"label":127,"unit":122,"statistic":123,"alignment":41},"Precision",{"label":129,"unit":122,"statistic":123,"alignment":41},"Recall",[131,134,136],{"dataset":132,"sequence":133,"environment":20},"simulated localizability dataset","Train",{"dataset":132,"sequence":135,"environment":20},"Valid",{"dataset":132,"sequence":137,"environment":138},"Test","simulation of mine tunnel",[140,143],{"name":141,"methodId":5,"linkable":142,"proposed":142,"self":142},"ResUNet (proposed)",true,{"name":144,"methodId":44,"linkable":66,"proposed":66,"self":66},"PointNet",[146,150,153,155,157,159,161,162,164,166,168,170,172,174,176,178,180,182,184,186,188,190,192,194],[147,147,147,148,149,147,149,149,147],0,0.998,-1,[147,147,151,152,149,147,149,149,147],1,0.961,[147,147,112,154,149,147,149,149,147],0.854,[147,151,147,156,149,147,149,149,147],0.997,[147,151,151,158,149,147,149,149,147],0.606,[147,151,112,160,149,147,149,149,147],0.517,[147,112,147,156,149,147,149,149,147],[147,112,151,163,149,147,149,149,147],0.647,[147,112,112,165,149,147,149,149,147],0.398,[147,167,147,148,149,147,149,149,147],3,[147,167,151,169,149,147,149,149,147],0.585,[147,167,112,171,149,147,149,149,147],0.752,[151,147,147,173,149,147,149,149,147],0.957,[151,147,151,175,149,147,149,149,147],0.945,[151,147,112,177,149,147,149,149,147],0.809,[151,151,147,179,149,147,149,149,147],0.94,[151,151,151,181,149,147,149,149,147],0.53,[151,151,112,183,149,147,149,149,147],0.214,[151,112,147,185,149,147,149,149,147],0.918,[151,112,151,187,149,147,149,149,147],0.472,[151,112,112,189,149,147,149,149,147],0.22,[151,167,147,191,149,147,149,149,147],0.959,[151,167,151,193,149,147,149,149,147],0.703,[151,167,112,195,149,147,149,149,147],0.244,[],[117],[],[],[201],"Localizability classification averaged over six dimensions; both networks trained 60 epochs on the same simulated splits; test set simulated from the tunnel ground-truth mesh",{"slug":203,"group":204,"sourceId":5,"sourceLabel":6,"table":205,"selfRows":112,"metrics":206,"seqs":210,"entrants":213,"cells":218,"outcomes":223,"locators":224,"hardware":225,"wordings":227,"notes":228},"nubert2022learninglocalizability-text-sec-v-b","nubert2022learninglocalizability:Text Sec. V-B","Text Sec. V-B",[207],{"label":208,"unit":209,"statistic":40,"alignment":41},"inference time","ms",[211],{"dataset":39,"sequence":212,"environment":39},"per scan",[214,216],{"name":215,"methodId":5,"linkable":142,"proposed":142,"self":142},"ResUNet (proposed), CPU-only",{"name":217,"methodId":5,"linkable":142,"proposed":142,"self":142},"ResUNet (proposed), GPU",[219,221],[147,147,147,220,149,147,147,149,147],28,[151,147,147,222,149,147,151,149,147],13,[],[102],[226,107],"Intel i7 10700F (CPU only)",[],[229],"Inference time of the PyTorch model inside a ROS node, no specific code optimization",[],1790510662058]