{"id":38655,"date":"2026-08-20T09:00:00","date_gmt":"2026-08-20T02:00:00","guid":{"rendered":"https:\/\/vmts.ch\/?p=38655"},"modified":"2026-08-20T10:07:50","modified_gmt":"2026-08-20T03:07:50","slug":"slam-scanner-accuracy-point-cloud-not-level","status":"publish","type":"post","link":"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/","title":{"rendered":"SLAM Scanner Accuracy: When the Point Cloud Is Not Level"},"content":{"rendered":"\n<div style=\"background: #000000; padding: 24px 28px; font-family: Arial,Helvetica,sans-serif; border-left: 4px solid #C70100; margin: 0 0 32px;\">\n<p style=\"font-size: 10px; font-weight: bold; letter-spacing: 0.16em; text-transform: uppercase; color: #c70100; margin: 0 0 14px;\">In short: 3 points<\/p>\n<ul style=\"list-style: none; padding: 0; margin: 0;\">\n<li style=\"display: flex; gap: 12px; padding: 8px 0; border-bottom: 0.5px solid #222222; align-items: flex-start;\"><span style=\"color: #c70100; font-weight: bold; font-size: 13px; flex-shrink: 0; margin-top: 1px; font-family: Arial,Helvetica,sans-serif;\">01<\/span>\n<p style=\"font-size: 13px; color: #cccccc; line-height: 1.55; margin: 0; font-family: Arial,Helvetica,sans-serif;\">The gap between a quoted 1 to 3 cm and the 3 to 5 cm we typically measure is not the main problem. The main problem is that the error is systematic: floors and ceilings come out tilted, not simply noisy. For modelling that means one thing: the tilt has to be fully removed before the first wall is drawn.<\/p>\n<\/li>\n<li style=\"display: flex; gap: 12px; padding: 8px 0; border-bottom: 0.5px solid #222222; align-items: flex-start;\"><span style=\"color: #c70100; font-weight: bold; font-size: 13px; flex-shrink: 0; margin-top: 1px; font-family: Arial,Helvetica,sans-serif;\">02<\/span>\n<p style=\"font-size: 13px; color: #cccccc; line-height: 1.55; margin: 0; font-family: Arial,Helvetica,sans-serif;\">An attitude error of 0.07 degrees, a figure one manufacturer publishes itself, produces roughly 2.4 cm of height difference across a 20 m floor plate. That is more than the flatness tolerance the slab was built to under DIN 18202.<\/p>\n<\/li>\n<li style=\"display: flex; gap: 12px; padding: 8px 0; align-items: flex-start;\"><span style=\"color: #c70100; font-weight: bold; font-size: 13px; flex-shrink: 0; margin-top: 1px; font-family: Arial,Helvetica,sans-serif;\">03<\/span>\n<p style=\"font-size: 13px; color: #cccccc; line-height: 1.55; margin: 0; font-family: Arial,Helvetica,sans-serif;\">You can test for this at a trade fair stand in ten minutes. Ask for a raw, ungeoreferenced scan of a long corridor, cut a section through the floor, and measure both ends. The full checklist is at the end of this article.<\/p>\n<\/li>\n<\/ul>\n<\/div>\n\n\n\n<p style=\"font-size: 16px; line-height: 1.7; color: #4a4a4a;\">INTERGEO opens in Munich on 15 September. If you have walked that floor in previous years you already know what to expect: another cohort of new scanner brands, more TLS and mobile mapping variants than the year before, LiDAR capture fused with Gaussian splatting for genuinely impressive visual output, published accuracy inside the range you need, and a price that can sit at around a fifth of an established brand. Every autumn our clients ask us the same question before they commit: is the SLAM scanner accuracy good enough for their projects? This article is our answer, based on the point clouds that actually reach our production floor.<\/p>\n\n\n\n<div class=\"wp-block-group\" style=\"border-left: 3px solid #DADADA; padding-left: 18px; margin-bottom: 20px;\"><div class=\"wp-block-group__inner-container is-layout-flow wp-block-group-is-layout-flow\">\n<p style=\"font-size: 15px; font-style: italic; line-height: 1.7; color: #4a4a4a;\"><strong>How accurate are low-cost SLAM scanners in practice?<\/strong> Instruments in this segment typically publish 1 to 3 cm. In the point clouds VMT Solutions processes, deviations of 3 to 5 cm are more common. The more important finding is that a large share of this error is systematic rather than random: horizontal surfaces such as floor slabs and ceilings are reconstructed tilted, so the deviation grows with the length of the building and cannot be averaged out by fitting a plane.<\/p>\n<\/div><\/div>\n\n\n\n<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_84 counter-hierarchy ez-toc-counter ez-toc-grey ez-toc-container-direction\">\n<div class=\"ez-toc-title-container\">\n<p class=\"ez-toc-title ez-toc-toggle\" style=\"cursor:pointer\">Table of Contents<\/p>\n<span class=\"ez-toc-title-toggle\"><a href=\"#\" class=\"ez-toc-pull-right ez-toc-btn ez-toc-btn-xs ez-toc-btn-default ez-toc-toggle\" aria-label=\"Toggle Table of Content\"><span class=\"ez-toc-js-icon-con\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">Toggle<\/span><span class=\"ez-toc-icon-toggle-span\"><svg style=\"fill: #999;color:#999\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" class=\"list-377408\" width=\"20px\" height=\"20px\" viewBox=\"0 0 24 24\" fill=\"none\"><path d=\"M6 6H4v2h2V6zm14 0H8v2h12V6zM4 11h2v2H4v-2zm16 0H8v2h12v-2zM4 16h2v2H4v-2zm16 0H8v2h12v-2z\" fill=\"currentColor\"><\/path><\/svg><svg style=\"fill: #999;color:#999\" class=\"arrow-unsorted-368013\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" width=\"10px\" height=\"10px\" viewBox=\"0 0 24 24\" version=\"1.2\" baseProfile=\"tiny\"><path d=\"M18.2 9.3l-6.2-6.3-6.2 6.3c-.2.2-.3.4-.3.7s.1.5.3.7c.2.2.4.3.7.3h11c.3 0 .5-.1.7-.3.2-.2.3-.5.3-.7s-.1-.5-.3-.7zM5.8 14.7l6.2 6.3 6.2-6.3c.2-.2.3-.5.3-.7s-.1-.5-.3-.7c-.2-.2-.4-.3-.7-.3h-11c-.3 0-.5.1-.7.3-.2.2-.3.5-.3.7s.1.5.3.7z\"\/><\/svg><\/span><\/span><\/span><\/a><\/span><\/div>\n<nav><ul class='ez-toc-list ez-toc-list-level-1 eztoc-toggle-hide-by-default' ><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#Why_clients_ask_us_and_not_the_vendor\" >Why clients ask us and not the vendor<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#The_number_on_the_datasheet_and_the_number_in_the_file\" >The number on the datasheet, and the number in the file<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-3\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#The_SLAM_scanner_accuracy_problem_that_is_not_on_the_datasheet_horizontal_planes_that_are_not_level\" >The SLAM scanner accuracy problem that is not on the datasheet: horizontal planes that are not level<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-4\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#Why_does_a_fraction_of_a_degree_become_five_centimetres\" >Why does a fraction of a degree become five centimetres?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-5\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#Why_does_a_tilted_floor_matter_more_than_the_number_suggests\" >Why does a tilted floor matter more than the number suggests?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-6\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#What_a_tilted_point_cloud_does_to_the_BIM_model\" >What a tilted point cloud does to the BIM model<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-7\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#The_tilt_has_to_be_corrected_before_modelling_starts\" >The tilt has to be corrected before modelling starts<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-8\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#The_other_five_things_we_flag_on_this_kind_of_dataset\" >The other five things we flag on this kind of dataset<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-9\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#How_can_you_test_SLAM_scanner_accuracy_before_you_buy\" >How can you test SLAM scanner accuracy before you buy?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-10\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#Where_these_instruments_genuinely_earn_their_price\" >Where these instruments genuinely earn their price<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-11\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#We_check_the_point_cloud_before_we_quote_not_after_we_model\" >We check the point cloud before we quote, not after we model<\/a><ul class='ez-toc-list-level-3' ><li class='ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-12\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#What_our_source_data_check_covers\" >What our source data check covers<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-13\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#VMTS_as_your_point_cloud_to_BIM_partner\" >VMTS as your point cloud to BIM partner<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-14\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#Frequently_asked_questions_about_SLAM_scanner_accuracy\" >Frequently asked questions about SLAM scanner accuracy<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-15\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#Considering_a_new_scanner_Send_us_a_test_dataset_first\" >Considering a new scanner? Send us a test dataset first<\/a><ul class='ez-toc-list-level-3' ><li class='ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-16\" href=\"https:\/\/vmts.ch\/en\/slam-scanner-accuracy-point-cloud-not-level\/#Sources\" >Sources<\/a><\/li><\/ul><\/li><\/ul><\/nav><\/div>\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Why_clients_ask_us_and_not_the_vendor\"><\/span>Why clients ask us and not the vendor<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>We do not sell hardware. We receive point clouds from every kind of instrument, from terrestrial laser scanners and handheld or backpack SLAM systems to mobile mapping platforms, photogrammetry and 360 degree capture devices, and we have to turn all of them into BIM models that a design team can build from. That position gives us an unusual view. We never see the demo dataset. We see the third project of the season, scanned in a hurry, in a building with long corridors and no GNSS.<\/p>\n\n\n\n<p>For that reason we do not publish device names alongside our findings, and we will not do so here. What follows describes a pattern we observe across a class of instruments, not a verdict on any single product. Several of these systems are well engineered and genuinely useful. The point of this article is that the specification sheet does not describe the failure mode you will actually meet.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"The_number_on_the_datasheet_and_the_number_in_the_file\"><\/span>The number on the datasheet, and the number in the file<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Low-cost SLAM instruments in this segment typically quote accuracy in the region of 1 to 3 cm. In the point clouds we process, the deviations we measure are more often in the 3 to 5 cm range. That gap is real, but on its own it is not disqualifying, and it is roughly what the independent literature would lead you to expect. A 2025 comparative study from the Czech Technical University measured RMSE after target based transformation of 1.2 mm for a Leica RTC360 and 5.4 mm for a Trimble X7, against 7.3 mm and 8.2 mm for two current generation SLAM systems and 35.3 mm for an older generation one. SLAM sits an order of magnitude behind static scanning, which everybody in the industry already accepts.<\/p>\n\n\n\n<p>The trouble is that a single accuracy figure implies a random error, meaning noise scattered evenly around the true surface, which a modeller can average out by fitting a plane. That is not what we find.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"The_SLAM_scanner_accuracy_problem_that_is_not_on_the_datasheet_horizontal_planes_that_are_not_level\"><\/span>The SLAM scanner accuracy problem that is not on the datasheet: horizontal planes that are not level<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<div class=\"wp-block-group\" style=\"border-left: 3px solid #DADADA; padding-left: 18px; margin-bottom: 20px;\"><div class=\"wp-block-group__inner-container is-layout-flow wp-block-group-is-layout-flow\">\n<p style=\"font-size: 15px; font-style: italic; line-height: 1.7; color: #4a4a4a;\">On the majority of the devices in this class, and on almost every project, the point cloud arrives tilted. Cut a vertical section through a floor slab or a ceiling and the surface does not read as a horizontal line with noise around it. It reads as a wedge: two surfaces that converge at one end of the building and diverge by several centimetres at the other.<\/p>\n<\/div><\/div>\n\n\n\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" class=\"lazyload wp-image-slam-tilt-5cm\" style=\"border-radius: 4px;\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27961%27%20height%3D%27665%27%20viewBox%3D%270%200%20961%20665%27%3E%3Crect%20width%3D%27961%27%20height%3D%27665%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/vmts.ch\/wp-content\/uploads\/2026\/08\/slam-scanner-accuracy-vertical-drift-5cm-section.png\" alt=\"SLAM scanner accuracy problem: vertical section through a floor slab showing a 5 cm height difference between two overlapping scan surfaces\" width=\"961\" height=\"665\" loading=\"lazy\" \/>\n<figcaption class=\"wp-element-caption\" style=\"color: #4a4a4a; font-size: 13px; font-style: italic;\">Section through a ceiling slab in a merged SLAM point cloud. The two surfaces should coincide. At this end of the building they are 5 cm apart (VMT Solutions, anonymised client data).<\/figcaption>\n<\/figure>\n\n\n\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" class=\"lazyload wp-image-slam-tilt-3cm\" style=\"border-radius: 4px;\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%27956%27%20height%3D%27632%27%20viewBox%3D%270%200%20956%20632%27%3E%3Crect%20width%3D%27956%27%20height%3D%27632%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/vmts.ch\/wp-content\/uploads\/2026\/08\/slam-scanner-accuracy-vertical-drift-3cm-section.png\" alt=\"Handheld SLAM scanner accuracy: the same slab measured at another position showing 3 cm vertical deviation between scan surfaces\" width=\"956\" height=\"632\" loading=\"lazy\" \/>\n<figcaption class=\"wp-element-caption\" style=\"color: #4a4a4a; font-size: 13px; font-style: italic;\">The same slab measured elsewhere in the building: 3 cm. The deviation is not constant. It varies with position, which is the signature of a tilt rather than an offset (VMT Solutions).<\/figcaption>\n<\/figure>\n\n\n\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" class=\"lazyload wp-image-slam-tilt-4cm\" style=\"border-radius: 4px;\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%271092%27%20height%3D%27609%27%20viewBox%3D%270%200%201092%20609%27%3E%3Crect%20width%3D%271092%27%20height%3D%27609%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/vmts.ch\/wp-content\/uploads\/2026\/08\/slam-scanner-accuracy-vertical-drift-4cm-section.png\" alt=\"SLAM drift in point cloud to BIM: 4 cm vertical separation between two scan passes over the same floor level\" width=\"1092\" height=\"609\" loading=\"lazy\" \/>\n<figcaption class=\"wp-element-caption\" style=\"color: #4a4a4a; font-size: 13px; font-style: italic;\">A third position, 4 cm. Across one storey of one building we measured 3, 4 and 5 cm on what is physically a single slab (VMT Solutions).<\/figcaption>\n<\/figure>\n\n\n\n<p>This is not our observation alone. A 2023 study in <em>Automation in Construction<\/em> by Keitaanniemi and colleagues at Aalto University analysed drift in indoor SLAM based laser scanning and found drift errors reaching 10.6 cm, reduced to 0.2 cm only after sectional non rigid post processing. Their key finding matches ours precisely: <strong>the drift error is mainly in the vertical direction, and it is largest in the middle of the trajectory<\/strong>.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Why_does_a_fraction_of_a_degree_become_five_centimetres\"><\/span>Why does a fraction of a degree become five centimetres?<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" class=\"lazyload wp-image-slam-attitude-error\" style=\"border-radius: 4px;\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%271672%27%20height%3D%27941%27%20viewBox%3D%270%200%201672%20941%27%3E%3Crect%20width%3D%271672%27%20height%3D%27941%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/vmts.ch\/wp-content\/uploads\/2026\/08\/slam-scanner-accuracy-attitude-error-0-07-degrees.jpg\" alt=\"SLAM scanner accuracy: an attitude error of 0.07 degrees produces 2.4 cm of height deviation across 20 m of building length in the point cloud\" width=\"1672\" height=\"941\" loading=\"lazy\" \/>\n<figcaption class=\"wp-element-caption\" style=\"color: #4a4a4a; font-size: 13px; font-style: italic;\">The lever arm is the building itself. Misjudging vertical by 0.07 degrees lifts the far end of the slab by 2.4 cm, while the laser measures distance perfectly well throughout (schematic, VMT Solutions).<\/figcaption>\n<\/figure>\n\n\n\n<p>The physics is not mysterious, and understanding it tells you what to test. A SLAM system estimates its own trajectory by matching each new observation against what it has already seen, in the manner of a surveying traverse. Small errors compound along the path. Two of the three rotation angles, roll and pitch, are in principle recoverable from the IMU, because gravity gives an absolute vertical reference. As Kubelka and colleagues put it in their 2022 work on gravity constrained registration, in the absence of constant acceleration the gravity vector makes two of the three attitude angles observable, reducing the problem from six degrees of freedom to four.<\/p>\n\n\n\n<p>In principle. In practice the levelling is only as good as the IMU and its calibration. One manufacturer&#8217;s own field guide states that the IMU alone contributes approximately 0.07 degrees of levelling error, and work with sub degree grade IMUs typically treats 0.1 degrees as the calibration threshold that matters. Those sound like negligible numbers until you carry them across a building:<\/p>\n\n\n\n<figure class=\"wp-block-table\">\n<table>\n<thead>\n<tr>\n<th>Attitude (levelling) error<\/th>\n<th>Height error over 10 m<\/th>\n<th>Over 20 m<\/th>\n<th>Over 40 m<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>0.05 degrees<\/td>\n<td>9 mm<\/td>\n<td>17 mm<\/td>\n<td>35 mm<\/td>\n<\/tr>\n<tr>\n<td><strong>0.07 degrees<\/strong> (published IMU figure)<\/td>\n<td><strong>12 mm<\/strong><\/td>\n<td><strong>24 mm<\/strong><\/td>\n<td><strong>49 mm<\/strong><\/td>\n<\/tr>\n<tr>\n<td>0.10 degrees<\/td>\n<td>17 mm<\/td>\n<td>35 mm<\/td>\n<td>70 mm<\/td>\n<\/tr>\n<tr>\n<td>0.15 degrees<\/td>\n<td>26 mm<\/td>\n<td>52 mm<\/td>\n<td>105 mm<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<figcaption class=\"wp-element-caption\" style=\"color: #4a4a4a; font-size: 13px; font-style: italic;\">Height difference across a horizontal span for a given tilt, calculated as span multiplied by tan(angle). A tenth of a degree is 3.5 cm across a 20 m floor plate.<\/figcaption>\n<\/figure>\n\n\n\n<p>That is the whole explanation for the 3 to 5 cm we measure. It is not a range accuracy problem. The laser is measuring distance perfectly well. The instrument simply does not know exactly which way is down, and every error in that estimate is multiplied by the length of the building.<\/p>\n\n\n\n<p>Three practical consequences follow. First, the error grows with building size, so a demo in a 10 m showroom will never reveal it. Second, it accumulates along the trajectory, so it is worst in long corridors, in stairwells, and in exactly the buildings where a mobile system looks most attractive. Third, and most importantly for us, <strong>it is a systematic error, not noise<\/strong>, and systematic errors cannot be averaged away.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Why_does_a_tilted_floor_matter_more_than_the_number_suggests\"><\/span>Why does a tilted floor matter more than the number suggests?<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Put 3 to 5 cm next to the tolerance the building itself was constructed to. DIN 18202, the German standard for tolerances in building construction, sets flatness limit deviations for finished floors (Table 3, row 3) of 12 mm over a 10 m measuring distance and 15 mm over 15 m. In Switzerland the parallel reference is SIA 414\/2:2016. The slab you are scanning was built flatter than the error your point cloud contains, by a factor of two to four.<\/p>\n\n\n\n<p>The consequence is not a rounding issue. It is that <strong>you can no longer distinguish an instrument error from a real building defect<\/strong>. When a section shows a floor dropping 4 cm across a room, that is either differential settlement, a sagging timber joist or a badly poured screed, all things a client pays a survey to discover, or it is the scanner&#8217;s attitude drift. From the point cloud alone there is no way to tell. For a renovation project, a structural assessment or a heritage building, that ambiguity removes most of the value of the survey.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"What_a_tilted_point_cloud_does_to_the_BIM_model\"><\/span>What a tilted point cloud does to the BIM model<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Up to this point the tilt is a measurement problem. The moment the point cloud becomes a BIM model, it turns into a design problem, and into the kind you can no longer see in the finished model.<\/p>\n\n\n\n<p>Modelling is done against the point cloud. Every storey height, every sill and lintel height, every slab thickness, every roof pitch and every spot height is taken from the surface where the scanner put it. If that surface is tilted, the model inherits the tilt element by element. The result then looks flawless: clean walls, clean storeys, a model that validates and exports. Switch the point cloud off and the error is invisible and no longer traceable.<\/p>\n\n\n\n<p>It is passed on regardless. Anyone who takes a height from the model, derives a section, calculates quantities or has a component manufactured to it carries that tilt forward without knowing. For detailed design in an existing building, a tilted point cloud is therefore not a quality shortcoming you can accept. It is a reason to stop until it has been corrected.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"The_tilt_has_to_be_corrected_before_modelling_starts\"><\/span>The tilt has to be corrected before modelling starts<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>A tilt is not a modeller&#8217;s task. It is a data problem, and it has to be solved before the first wall is drawn, not alongside the modelling and not afterwards. Discovered in a finished model, it does not call for a correction. It calls for remodelling.<\/p>\n\n\n\n<p>It is easiest to remove while the individual scan positions or trajectory sections still exist. The Aalto study reduced 10.6 cm to 0.2 cm precisely by cutting the trajectory into roughly 30 second sections and applying a non rigid transformation to each. Control points measured with a total station achieve the same result, which is why every serious mobile mapping workflow includes them.<\/p>\n\n\n\n<p>What arrives on our desk, however, is very often a single merged E57 or LAS file with no trajectory, no scan boundaries and no control points. At that point standard tools run out. There is no rigid transformation that levels one end of the building without un levelling the other, and there is no automatic function that does the job for you.<\/p>\n\n\n\n<p>What remains is manual correction: specialised software, a process defined for exactly this case, and Geomatiker who have worked through enough datasets of the same kind. The real difficulty is not the arithmetic, it is the judgement. Which surface in the building is sound enough to serve as a height reference, and which one is already deformed? Where does instrument drift end and real settlement begin? That distinction cannot be automated, and it decides whether the corrected point cloud is dependable or merely wrong in a different way. At the end of that work you get a levelled point cloud you can model on, and a record of what was corrected and what the correction was referenced to.<\/p>\n\n\n\n<div class=\"wp-block-group has-background\" style=\"background-color: #f5f5f5; border-left: 3px solid #C70100; padding: 20px 24px 20px 24px;\"><div class=\"wp-block-group__inner-container is-layout-flow wp-block-group-is-layout-flow\">\n<p style=\"font-size: 10px; font-style: normal; font-weight: bold; letter-spacing: 0.12em; text-transform: uppercase; color: #c70100;\">VMTS practice note<\/p>\n\n\n\n<p style=\"font-size: 20px; font-weight: bold; line-height: 1.3; color: #000000;\">Always send us the individual scans or trajectory sections, not only the merged cloud.<\/p>\n\n\n\n<p style=\"color: #4a4a4a;\">Keeping the raw project alongside the export costs you nothing at the time of survey and preserves every option later. It lets us identify a single divergent scan and remove it, correct a tilt section by section, and tell you which deviations are the building and which are the instrument. Merged files are convenient to send and expensive to repair.<\/p>\n<\/div><\/div>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"The_other_five_things_we_flag_on_this_kind_of_dataset\"><\/span>The other five things we flag on this kind of dataset<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>The tilt is the one that concerns us most, because it is systematic and largely invisible in a viewer. The others are more mundane, but each of them costs modelling time if it is not caught before work starts. This is the list our Geomatiker work through on incoming data:<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>A single divergent scan inside the merged set.<\/strong> One position registers several centimetres away from the rest and drags the local geometry with it. In a merged file it is often easier to discard than to correct, provided you can still identify which points belong to it.<\/li>\n<li><strong>The 0.00 reference does not sit on the ground floor.<\/strong> The project zero ends up on an arbitrary level, so every storey and every dimension in the model has to be re based before the client can use it.<\/li>\n<li><strong>The point cloud sits far from the coordinate origin.<\/strong> Real world national coordinates, applied directly, place the geometry tens of thousands of metres from origin. Most BIM authoring tools lose single precision accuracy at that distance and produce rounding and snapping errors during modelling.<\/li>\n<li><strong>No coverage above the roof.<\/strong> Roof build up thickness then has to be inferred from the visible underside, which is an assumption in the model rather than a measurement, and it must be declared as such.<\/li>\n<li><strong>No reference photography.<\/strong> Without site images, ambiguous point cloud regions cannot be resolved and the modeller has to guess. We ask for photographs on every project for this reason.<\/li>\n<\/ol>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"How_can_you_test_SLAM_scanner_accuracy_before_you_buy\"><\/span>How can you test SLAM scanner accuracy before you buy?<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>None of this means you should not buy one. It means the demo you are shown at a stand is not the test that answers your question. Six things to ask for, all of which a confident vendor will agree to:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Ask for the raw, ungeoreferenced dataset.<\/strong> Not the polished sample, not the RTK corrected version, and not a Gaussian splatting render. Visual fidelity and geometric accuracy are independent properties, and a splat can look photographic while the underlying geometry drifts. Ask for the point cloud the SLAM engine produced on its own.<\/li>\n<li><strong>Ask what the accuracy figure refers to.<\/strong> Most quoted figures describe <em>relative<\/em> accuracy, meaning internal consistency over a stated distance, often on the order of 1 cm within 100 m. Absolute accuracy, the number that matters when the model has to match a site survey, generally requires RTK or ground control and is a different figure. If the datasheet does not distinguish them, ask which one is printed.<\/li>\n<li><strong>Test the levelness directly.<\/strong> Take a scan of a long corridor or a large floor plate, cut a vertical section through the floor slab, and measure the height at both ends. Any building over about 20 m will expose an attitude error of a tenth of a degree.<\/li>\n<li><strong>Test without a loop closure.<\/strong> Walk out and back along the same route without returning to the start from a different direction, which is the case where loop closure typically fails to trigger. This is also the normal situation in a linear building.<\/li>\n<li><strong>Scan the same space twice and compare the two clouds.<\/strong> Repeatability is the cheapest proxy for systematic error, and it needs no reference data at all. If two passes of the same corridor disagree by 4 cm, you have your answer.<\/li>\n<li><strong>Ask what the export contains.<\/strong> Trajectory, individual scan sections, control point residuals, a quality report. If the only deliverable is a merged cloud, every downstream correction option is closed to you before you start.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Where_these_instruments_genuinely_earn_their_price\"><\/span>Where these instruments genuinely earn their price<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>It would be dishonest to end on the negative. A systematic vertical drift matters enormously for some deliverables and hardly at all for others, and the difference is worth knowing before you write off a device that costs a fifth of the alternative.<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" class=\"lazyload wp-image-slam-suitability-lod\" style=\"border-radius: 4px;\" src=\"data:image\/svg+xml,%3Csvg%20xmlns%3D%27http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%27%20width%3D%271672%27%20height%3D%27941%27%20viewBox%3D%270%200%201672%20941%27%3E%3Crect%20width%3D%271672%27%20height%3D%27941%27%20fill-opacity%3D%220%22%2F%3E%3C%2Fsvg%3E\" data-orig-src=\"https:\/\/vmts.ch\/wp-content\/uploads\/2026\/08\/slam-suitability-lod-100-200-vs-lod-300-precision.jpg\" alt=\"When SLAM is good enough and when it is not: LOD 100 to 200 tolerates 3 to 5 cm of drift, while LOD 300 requires control points and 15 mm dimensional accuracy\" width=\"1672\" height=\"941\" loading=\"lazy\" \/>\n<figcaption class=\"wp-element-caption\" style=\"color: #4a4a4a; font-size: 13px; font-style: italic;\">The deliverable decides, not the instrument. For area take off, inventory documentation and LOD 100 to 200, a 3 to 5 cm drift changes nothing. From LOD 300 and in prefabrication, the governing figure is &plusmn;15 mm, which only control point supported capture can carry (VMT Solutions).<\/figcaption>\n<\/figure>\n\n\n\n<p>Where the error is largely irrelevant: volumetric and area take off, stock and inventory documentation, visual site records and progress documentation, clash detection against generous tolerances, feasibility studies, and LOD 100 to LOD 200 modelling where a few centimetres over a facade does not change a decision. In these applications the productivity gain of walking a building in twenty minutes is real and substantial, and we process this data routinely with good results.<\/p>\n\n\n\n<p>Where it is not acceptable: LOD 300 detailed design, where our documented dimensional accuracy of &plusmn;15 mm is smaller than the drift itself; prefabrication and fit out, where components are manufactured to the model; structural and settlement assessment, where the whole point is to measure real deviation; heritage documentation; and any deliverable that has to tie into a site survey or a national coordinate system. In those cases, either use a static scanner, or use the mobile system with control points and keep the trajectory sections.<\/p>\n\n\n\n<p>The instrument is not the problem. Buying it on a specification that does not describe its dominant error mode is.<\/p>\n\n\n\n<div class=\"wp-block-group has-background\" style=\"background-color: #f5f5f5; border-left: 4px solid #C70100; padding: 32px;\"><div class=\"wp-block-group__inner-container is-layout-flow wp-block-group-is-layout-flow\">\n\n\n<p style=\"font-size: 10px; font-style: normal; font-weight: bold; letter-spacing: 0.16em; text-transform: uppercase; color: #c70100; margin: 0 0 12px;\">The VMTS difference<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" style=\"margin-top: 0;\"><span class=\"ez-toc-section\" id=\"We_check_the_point_cloud_before_we_quote_not_after_we_model\"><\/span>We check the point cloud before we quote, not after we model<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Every dataset that reaches us goes through a source data check by our Geomatiker before a single wall is drawn. We section the horizontal planes and measure them, we look for divergent scan positions, we verify the coordinate system and the reference level, and we report the gaps back to you in writing. If a point cloud cannot support the level of detail you have asked for, you learn that at quotation stage, while you can still decide to rescan, rather than at handover when the only remaining option is to argue about it.<\/p>\n\n\n\n<p>This is why our clients send us their point cloud for checking before they work with it any further. Checking the point cloud and, where necessary, levelling it are a fixed step before a project starts at VMTS, not a special case. What goes into modelling afterwards is a point cloud whose condition has been verified, corrected and documented in writing.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" style=\"margin-top: 20px;\"><span class=\"ez-toc-section\" id=\"What_our_source_data_check_covers\"><\/span>What our source data check covers<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>Levelness checked and restored:<\/strong> sections through floor and ceiling slabs at several positions, deviations measured and recorded, tilt corrected before modelling begins<\/li>\n<li><strong>Registration consistency:<\/strong> identification of individual scans that diverge from the set, with a recommendation to correct or exclude<\/li>\n<li><strong>Coordinate system and origin distance:<\/strong> verification of the reference system and repositioning close to origin for modelling, with the survey position restored at export<\/li>\n<li><strong>Reference level:<\/strong> confirmation that 0.00 sits where the project expects it, before storeys are built<\/li>\n<li><strong>Coverage and completeness:<\/strong> a written list of areas without data, so that assumptions in the model are declared as assumptions<\/li>\n<li><strong>QC report at handover:<\/strong> delivered with the native file and IFC2x3, listing every check performed and every deviation found<\/li>\n<\/ol>\n\n\n\n<p style=\"margin: 0;\">This is our guiding principle in practice: &#8220;You scan, we support all downstream services, from point clouds to native BIM models.&#8221; You choose the instrument that fits your business. We tell you honestly what its data can and cannot support, and we deliver the model at the level of detail agreed before the project starts.<\/p>\n\n\n<\/div><\/div>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"VMTS_as_your_point_cloud_to_BIM_partner\"><\/span>VMTS as your point cloud to BIM partner<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>VMT Solutions has been converting point cloud data into native BIM models since 2009, in Archicad, Revit, Allplan and Vectorworks, for surveying firms, architectural practices and building owners across the DACH region and worldwide. We process seven point cloud formats, E57, RCP, LAS, FLS, LAZ, XYZ and PTX, from every class of instrument, and we deliver IFC2x3 and a QC report alongside the native file.<\/p>\n\n\n\n<div style=\"display: grid; grid-template-columns: 1fr 1fr 1fr; gap: 1px; background: #DADADA; border: 0.5px solid #DADADA; font-family: Arial,Helvetica,sans-serif; margin: 0 0 24px;\">\n<div style=\"background: #ffffff; padding: 24px 20px; text-align: center;\">\n<div style=\"font-size: 36px; font-weight: bold; color: #c70100; line-height: 1;\">3&ndash;5 cm<\/div>\n<div style=\"font-size: 11px; color: #9e9e9e; margin-top: 6px; text-transform: uppercase; letter-spacing: 0.08em; line-height: 1.4;\">Typical tilt we measure in this device class<\/div>\n<\/div>\n<div style=\"background: #ffffff; padding: 24px 20px; text-align: center; border-left: 0.5px solid #DADADA; border-right: 0.5px solid #DADADA;\">\n<div style=\"font-size: 36px; font-weight: bold; color: #000000; line-height: 1;\">&plusmn;15 mm<\/div>\n<div style=\"font-size: 11px; color: #9e9e9e; margin-top: 6px; text-transform: uppercase; letter-spacing: 0.08em; line-height: 1.4;\">Dimensional accuracy from LOD 300<\/div>\n<\/div>\n<div style=\"background: #ffffff; padding: 24px 20px; text-align: center;\">\n<div style=\"font-size: 36px; font-weight: bold; color: #000000; line-height: 1;\">2500+<\/div>\n<div style=\"font-size: 11px; color: #9e9e9e; margin-top: 6px; text-transform: uppercase; letter-spacing: 0.08em; line-height: 1.4;\">Completed projects worldwide<\/div>\n<\/div>\n<\/div>\n\n\n\n<p>Read more about how we define scope in our article on <a href=\"https:\/\/vmts.ch\/en\/what-is-lod-200-300-400-in-bim\/\">LOD 100 to 400 in BIM<\/a>, about capture method and accuracy in <a href=\"https:\/\/vmts.ch\/en\/is-matterport-data-accurate-enough-for-bim\/\">Matterport data accuracy for BIM<\/a>, see how our <a href=\"https:\/\/vmts.ch\/en\/people-behind-bim-accuracy-vmt-solutions\/\">Geomatiker verify the point cloud<\/a> before modelling begins, or review our <a href=\"https:\/\/vmts.ch\/en\/scan-to-bim\/\">Scan to BIM services<\/a>.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Frequently_asked_questions_about_SLAM_scanner_accuracy\"><\/span>Frequently asked questions about SLAM scanner accuracy<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\">\n<summary>How accurate are low-cost SLAM scanners in practice?<\/summary>\n\n<p>Devices in this segment typically quote 1 to 3 cm. In the point clouds we process, deviations of 3 to 5 cm are more common. Independent testing supports the order of magnitude: a 2025 study measured RMSE of 7.3 to 8.2 mm for current generation SLAM systems and 35.3 mm for an older generation one, against 1.2 to 5.4 mm for static scanners. The more important point is that a large part of the SLAM error is systematic rather than random.<\/p>\n<\/details>\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\">\n<summary>Why is my point cloud tilted or not level?<\/summary>\n\n<p>Because the system&#8217;s estimate of vertical is imperfect. Roll and pitch come from the IMU&#8217;s reading of gravity, and any error in that estimate is multiplied by the length of the building. An attitude error of 0.07 degrees produces about 2.4 cm of height difference across a 20 m span, and 0.1 degrees produces 3.5 cm. The effect accumulates along the trajectory, so it is worst in long corridors and largest in the middle of the scan route.<\/p>\n<\/details>\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\">\n<summary>What happens if I model on a tilted point cloud?<\/summary>\n\n<p>The model inherits the tilt element by element. Storey heights, sill and lintel heights, slab thicknesses, roof pitches and spot heights are all taken from a surface that does not sit where it sits in the building. The finished model still looks flawless, because with the point cloud switched off the error is no longer visible. It is passed on regardless, to every height figure, every section, every quantity take off and every component manufactured to the model. That is why the tilt has to be corrected before modelling starts, not after.<\/p>\n<\/details>\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\">\n<summary>Can a tilted point cloud be corrected afterwards?<\/summary>\n\n<p>Yes, but not automatically. If the individual scans or trajectory sections still exist, the correction is comparatively direct: published research has reduced a 10.6 cm drift to 0.2 cm using sectional non rigid transformation, and control points measured with a total station achieve the same. If only a merged cloud remains, with no trajectory and no control points, no rigid transformation helps, because it can only level one end at the expense of the other. What is left is manual correction with specialised software, a defined process and experienced Geomatiker who can judge which surface in the building is sound enough to serve as a height reference. VMT Solutions carries out this levelling as a fixed step before modelling. Keep the raw project alongside the export anyway, because it shortens the work considerably.<\/p>\n<\/details>\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\">\n<summary>Is a SLAM point cloud good enough for point cloud to BIM?<\/summary>\n\n<p>It depends on the level of detail. For LOD 100 to LOD 200, volumetric take off, inventory documentation and visual records, current SLAM data is usually more than adequate and the productivity gain is substantial. For LOD 300 detailed design, prefabrication, structural or settlement assessment and heritage documentation, a 3 to 5 cm systematic error is larger than the tolerance the work requires, and either a static scanner or control point supported mobile capture is needed.<\/p>\n<\/details>\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\">\n<summary>What is the difference between relative and absolute accuracy on a scanner datasheet?<\/summary>\n\n<p>Relative accuracy describes internal consistency within the point cloud over a stated distance, for example 1 cm within 100 m, and can be achieved by scanning technique alone. Absolute accuracy describes how well the cloud matches real world coordinates, and generally requires RTK or ground control points. Most headline figures in this segment are relative. If your model has to tie into a site survey, absolute accuracy is the figure that governs.<\/p>\n<\/details>\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\">\n<summary>Does Gaussian splatting improve scanning accuracy?<\/summary>\n\n<p>No. Gaussian splatting is a rendering and reconstruction technique that produces highly convincing visual output. It affects how the capture looks, not how well the instrument knows where it is. A dataset can look photographic and still carry several centimetres of trajectory drift. When evaluating an instrument, assess the geometry separately from the visualisation.<\/p>\n<\/details>\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\">\n<summary>How can I test a scanner at a trade fair?<\/summary>\n\n<p>Ask for the raw, ungeoreferenced point cloud of a long corridor or large floor plate, scanned without a loop closure. Cut a vertical section through the floor slab and measure the height at both ends. Then scan the same space a second time and compare the two clouds. Any building over roughly 20 m will expose an attitude error of a tenth of a degree, and repeatability between two passes is the cheapest available proxy for systematic error.<\/p>\n<\/details>\n\n\n<div class=\"wp-block-group has-background\" style=\"background-color: #000000; padding: 48px;\"><div class=\"wp-block-group__inner-container is-layout-flow wp-block-group-is-layout-flow\">\n<p style=\"font-size: 10px; font-style: normal; font-weight: bold; letter-spacing: 0.16em; text-transform: uppercase; color: #c70100;\">Work with VMTS<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" style=\"color: #ffffff; font-size: 24px; font-weight: 400; line-height: 1.3; border: none;\"><span class=\"ez-toc-section\" id=\"Considering_a_new_scanner_Send_us_a_test_dataset_first\"><\/span>Considering a new scanner? Send us a test dataset first<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p style=\"color: #888888; font-size: 14px;\">Before you commit to an instrument, send us a raw scan from it. We will run our source data check, section the horizontal planes, measure the deviations and send you the findings in writing, usually within 24 hours and with no obligation. We do not sell hardware, so you get an assessment of the data rather than of the brand. The same applies to projects already running: send us the point cloud before it is modelled. We would be happy to discuss the result with you.<\/p>\n\n\n\n<div class=\"wp-block-buttons is-layout-flex wp-block-buttons-is-layout-flex\">\n<div class=\"wp-block-button\"><a class=\"fusion-button button-flat button-large button-custom fusion-button-default fusion-button-default-span fusion-button-default-type\" style=\"--button_accent_color: var(--awb-color1); --button_accent_hover_color: var(--awb-color1); --button_border_hover_color: var(--awb-color1); --button_gradient_top_color: var(--awb-custom_color_1); --button_gradient_bottom_color: var(--awb-custom_color_1); --button_gradient_top_color_hover: hsla(var(--awb-custom_color_1-h),var(--awb-custom_color_1-s),calc(var(--awb-custom_color_1-l) - 12%),var(--awb-custom_color_1-a)); --button_gradient_bottom_color_hover: hsla(var(--awb-custom_color_1-h),var(--awb-custom_color_1-s),calc(var(--awb-custom_color_1-l) - 12%),var(--awb-custom_color_1-a)); --button_text_transform: uppercase;\" href=\"https:\/\/vmts.ch\/en\/contact-us\/\" target=\"_self\" data-hover=\"icon_slide\"><span class=\"fusion-button-text awb-button__text awb-button__text--default\">Send us a test point cloud<\/span><i class=\"fa-long-arrow-alt-right fas awb-button__icon awb-button__icon--default button-icon-right\" aria-hidden=\"true\"><\/i><\/a><\/div>\n<\/div>\n<\/div><\/div>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Sources\"><\/span>Sources<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Keitaanniemi, A., R&ouml;nnholm, P., Kukko, A., Vaaja, M. T. (2023): <em>Drift analysis and sectional post-processing of indoor simultaneous localization and mapping (SLAM)-based laser scanning data<\/em>. Automation in Construction. Aalto University.<\/li>\n<li>Braun, J. et al. (2025): <em>A Comparative Study of Indoor Accuracies Between SLAM and Static Scanners<\/em>. Applied Sciences 15(14), 8053. Czech Technical University in Prague.<\/li>\n<li>Kubelka, V. et al. (2022): <em>Gravity-constrained point cloud registration<\/em>. arXiv:2203.13799.<\/li>\n<li>DIN 18202 <em>Toleranzen im Hochbau, Bauwerke<\/em>, Table 3 (Ebenheitstoleranzen), row 3, finished floors. Swiss parallel reference: SIA 414\/2:2016.<\/li>\n<li>Manufacturer field documentation on SLAM drift, loop closure and relative versus absolute accuracy (device name withheld).<\/li>\n<\/ul>\n\n\n\n<p><script type=\"application\/ld+json\">{\"@context\":\"https:\/\/schema.org\",\"@type\":\"TechArticle\",\"headline\":\"SLAM Scanner Accuracy: When the Point Cloud Is Not Level\",\"description\":\"Low-cost SLAM scanners quote 1 to 3 cm. We measure 3 to 5 cm and a systematic tilt. What to test before you buy, from a point cloud to BIM team.\",\"inLanguage\":\"en\",\"datePublished\":\"2026-08-20\",\"dateModified\":\"2026-08-20\",\"author\":{\"@type\":\"Organization\",\"name\":\"VMT Solutions Co., LTD\",\"url\":\"https:\/\/vmts.ch\/\"},\"publisher\":{\"@type\":\"Organization\",\"name\":\"VMT Solutions Co., LTD\",\"url\":\"https:\/\/vmts.ch\/\"},\"about\":[{\"@type\":\"Thing\",\"name\":\"SLAM scanner accuracy\"},{\"@type\":\"Thing\",\"name\":\"Point cloud to BIM\"},{\"@type\":\"Thing\",\"name\":\"Simultaneous localization and mapping\"},{\"@type\":\"Thing\",\"name\":\"Laser scanning\"},{\"@type\":\"Thing\",\"name\":\"Point cloud correction\"}],\"citation\":[\"Keitaanniemi, A., Ronnholm, P., Kukko, A., Vaaja, M. T. (2023). Drift analysis and sectional post-processing of indoor SLAM-based laser scanning data. Automation in Construction.\",\"Braun, J. et al. (2025). A Comparative Study of Indoor Accuracies Between SLAM and Static Scanners. Applied Sciences 15(14), 8053.\",\"Kubelka, V. et al. (2022). Gravity-constrained point cloud registration. arXiv:2203.13799.\",\"DIN 18202 Toleranzen im Hochbau, Table 3, row 3.\"]}<\/script><\/p>\n\n\n\n<p><script type=\"application\/ld+json\">{\"@context\":\"https:\/\/schema.org\",\"@type\":\"FAQPage\",\"mainEntity\":[{\"@type\":\"Question\",\"name\":\"How accurate are low-cost SLAM scanners in practice?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"Devices in this segment typically quote 1 to 3 cm. In the point clouds we process, deviations of 3 to 5 cm are more common. Independent testing supports the order of magnitude: a 2025 study measured RMSE of 7.3 to 8.2 mm for current generation SLAM systems and 35.3 mm for an older generation one, against 1.2 to 5.4 mm for static scanners. The more important point is that a large part of the SLAM error is systematic rather than random.\"}},{\"@type\":\"Question\",\"name\":\"Why is my point cloud tilted or not level?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"Because the system's estimate of vertical is imperfect. Roll and pitch come from the IMU's reading of gravity, and any error in that estimate is multiplied by the length of the building. An attitude error of 0.07 degrees produces about 2.4 cm of height difference across a 20 m span, and 0.1 degrees produces 3.5 cm. The effect accumulates along the trajectory, so it is worst in long corridors and largest in the middle of the scan route.\"}},{\"@type\":\"Question\",\"name\":\"What happens if I model on a tilted point cloud?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"The model inherits the tilt element by element. Storey heights, sill and lintel heights, slab thicknesses, roof pitches and spot heights are all taken from a surface that does not sit where it sits in the building. The finished model still looks flawless, because with the point cloud switched off the error is no longer visible. It is passed on regardless, to every height figure, every section, every quantity take off and every component manufactured to the model. That is why the tilt has to be corrected before modelling starts.\"}},{\"@type\":\"Question\",\"name\":\"Can a tilted point cloud be corrected afterwards?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"Yes, but not automatically. If the individual scans or trajectory sections still exist, the correction is comparatively direct: published research has reduced a 10.6 cm drift to 0.2 cm using sectional non rigid transformation, and control points measured with a total station achieve the same. If only a merged cloud remains, with no trajectory and no control points, no rigid transformation helps, because it can only level one end at the expense of the other. What is left is manual correction with specialised software, a defined process and experienced Geomatiker. VMT Solutions carries out this levelling as a fixed step before modelling.\"}},{\"@type\":\"Question\",\"name\":\"Is a SLAM point cloud good enough for point cloud to BIM?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"It depends on the level of detail. For LOD 100 to LOD 200, volumetric take off, inventory documentation and visual records, current SLAM data is usually more than adequate. For LOD 300 detailed design, prefabrication, structural or settlement assessment and heritage documentation, a 3 to 5 cm systematic error is larger than the tolerance the work requires, and either a static scanner or control point supported mobile capture is needed.\"}},{\"@type\":\"Question\",\"name\":\"What is the difference between relative and absolute accuracy on a scanner datasheet?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"Relative accuracy describes internal consistency within the point cloud over a stated distance, for example 1 cm within 100 m, and can be achieved by scanning technique alone. Absolute accuracy describes how well the cloud matches real world coordinates, and generally requires RTK or ground control points. Most headline figures in this segment are relative.\"}},{\"@type\":\"Question\",\"name\":\"Does Gaussian splatting improve scanning accuracy?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"No. Gaussian splatting is a rendering and reconstruction technique that produces highly convincing visual output. It affects how the capture looks, not how well the instrument knows where it is. A dataset can look photographic and still carry several centimetres of trajectory drift.\"}},{\"@type\":\"Question\",\"name\":\"How can I test a scanner at a trade fair?\",\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"Ask for the raw, ungeoreferenced point cloud of a long corridor or large floor plate, scanned without a loop closure. Cut a vertical section through the floor slab and measure the height at both ends. Then scan the same space a second time and compare the two clouds. Any building over roughly 20 m will expose an attitude error of a tenth of a degree.\"}}]}<\/script><\/p>\n\n","protected":false},"excerpt":{"rendered":"<p>In short: 3 points 01 The gap between a quoted 1 to 3 cm and the 3 to 5 cm we typically measure is not the main problem. The main problem is that the error [&#8230;]<\/p>\n","protected":false},"author":0,"featured_media":38649,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[325],"tags":[],"class_list":["post-38655","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-point-cloud-to-bim"],"acf":[],"_links":{"self":[{"href":"https:\/\/vmts.ch\/en\/wp-json\/wp\/v2\/posts\/38655","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/vmts.ch\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/vmts.ch\/en\/wp-json\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/vmts.ch\/en\/wp-json\/wp\/v2\/comments?post=38655"}],"version-history":[{"count":0,"href":"https:\/\/vmts.ch\/en\/wp-json\/wp\/v2\/posts\/38655\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/vmts.ch\/en\/wp-json\/wp\/v2\/media\/38649"}],"wp:attachment":[{"href":"https:\/\/vmts.ch\/en\/wp-json\/wp\/v2\/media?parent=38655"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/vmts.ch\/en\/wp-json\/wp\/v2\/categories?post=38655"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/vmts.ch\/en\/wp-json\/wp\/v2\/tags?post=38655"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}