YT YouTube2Text

YouTube2Text → Transcripts → SLAM WARS '26 : 9 Handheld LiDAR Scanner Competition (NavVis vs CHCNav vs FJD vs GreenValley & more)

SLAM WARS '26 : 9 Handheld LiDAR Scanner Competition (NavVis vs CHCNav vs FJD vs GreenValley & more) — Transcript

by The 3rd Dimension · 6,290 words · 899 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:15Yes.
0:1610 laser scanners for a head-to-head
0:19comparison, all capturing data in the
0:22same area using [music] the same control
0:24and under the same conditions. It
0:26actually did a really good job with max
0:28residuals in each of the northing,
0:30easting, and vertical axes of 11, 8, and
0:3410 mm. It's almost impressive how bad
0:37this scanner is doing to be honest. This
0:39is essentially useless data, nothing but
0:43noise. I was very pleasantly surprised
0:46with this scanner. And this was a
0:47competition after all, not everyone can
0:49win.
0:55When a surveyor is shopping for a total
0:57station, there's a very easy way to
0:59compare the accuracy of one unit to
1:01another. ISO, or the International
1:03Organization for Standardization, has
1:06created strict testing guidelines each
1:09manufacturer can follow to determine the
1:12accuracy of the various measurements
1:14their instrument makes. This way, when
1:16I'm looking at the spec sheet of a
1:18Trimble S7, I can directly compare it
1:20apples-to-apples to a Leica TS20. And I
1:23can even read and follow the ISO specs
1:26that govern the angular or distance
1:28accuracy to perform my own testing and
1:31ensure the instrument is operating
1:33within its given specification. And to
1:35take it a step further, I can even use
1:37error propagation to statistically
1:40estimate the accuracy of a derived point
1:42coordinate based on those ISO
1:45specifications, among other variables.
1:47In the world of laser scanning though,
1:49we don't have this luxury. There are no
1:51ISO specs or equivalent that are widely
1:54used and compare the range or angular
1:57accuracy of a 3D scanner. ISO 17123-9
2:02does cover terrestrial scanners, but not
2:05at the same individual component level
2:07that we get from a total station. And
2:09forget trying to use air propagation to
2:12get a sense of the accuracy of a point
2:14cloud captured from a handheld scanner.
2:17Because of this, we often get wild
2:19accuracy claims from manufacturers that
2:22are void of any kind of standardization
2:24or even any kind of explanation as to
2:27how they were measured in the first
2:28place, let alone any kind of evidence
2:31backing them up. Instead of apples to
2:33apples, we're comparing unicorns to
2:35leprechauns. So, when it comes time for
2:37Joe Surveyor to purchase his next laser
2:40scanner, this all makes his ability to
2:42make an educated decision rather
2:44difficult. And to place any kind of
2:46trust on the accuracy claims of the
2:48scanners that manufacturers make is, in
2:51my opinion, ill-advised. They can give
2:54us accuracy specs of the lidar sensor,
2:56which are problematic themselves, but to
2:58extrapolate that to the accuracy of a
3:01derived point cloud is akin to flinging
3:04at a wall and hoping it sticks. So,
3:06what does Joe do? Well, if I was in his
3:09shoes, I would love to see a bunch of
3:11similar laser scanners all capturing
3:13data in the same area using the same
3:16control and under the same conditions.
3:19And then have those data sets made
3:20publicly available to me so I could pick
3:22them apart and run my own analysis or
3:25have someone that's truly unbiased
3:28and has experience, tools, and the
3:30knowledge to do that and package it up
3:32in an easy-to-watch video. If you're not
3:35Joe, I got your back. Today, I'm
3:37unveiling the 2026 Third Dimension Slam
3:41Off. When I was at Intergeo this year, I
3:44invited seven manufacturers out to a
3:46testing area that was designed to
3:48simulate the various geometry types that
3:51one would come across during typical
3:53infield scan data acquisition. Those
3:56manufacturers brought a total of 10
3:59laser scanners for a head-to-head
4:01comparison to see who could create a
4:04point cloud that had the most utility,
4:06who had the best colorization, who had
4:08the highest degree of relative accuracy,
4:11and who had the best point cloud
4:12fidelity that preserved complex
4:15geometry. They all had the same rules
4:17for data capture, the same area to
4:19capture, and access to the same control.
4:22And most importantly, since I have no
4:24direct affiliation with any of these
4:26companies, you're all getting a truly
4:29unbiased review of the point clouds, and
4:32those data sets are going to be made
4:34available for you to download and run
4:36your own analysis on via the download
4:39link in the description of this video.
4:41100% transparency without bias. That's
4:45what you're getting today. And not to
4:47spoil the surprise, but the difference
4:49in quality between data sets was
4:51shocking, and the fact that most of
4:53these scanners share very similar
4:55hardware makes it even more
4:57mind-blowing. Before we dive into this,
4:59I need to hit you all with a disclaimer
5:01about what this test was and what it
5:04wasn't, what the rules were, and how I'm
5:06comparing data sets. This all took place
5:09at the Experience Park hosted by
5:10Navisate Intergeo this year. The total
5:13site was approximately 1,000 square
5:15meters or 10,000 square feet. Each team
5:18performing the scan was first walked
5:20through the area of interest, shown the
5:22location of the control points, what
5:25they had to scan, and told that they
5:27could capture the site in any manner
5:29they felt would provide the best
5:31results. No time limit was given, and I
5:33left full control of the capture and
5:36processing in their hands. This was an
5:39active site with other people around, so
5:41I made sure to encourage all
5:42participants to come back later in the
5:45evening when most conference goers had
5:47left if they felt this would improve the
5:49results. All control points were
5:51demarcated by adhesive targets on the
5:54ground or walls whose coordinates were
5:56defined by a least squares adjusted
5:59total station control network with
6:01coordinate standard deviations in the
6:03range of a couple millimeters. And a
6:05quick shout-out to Roth booker for
6:07lending me a four stand and paddle for
6:09the terrestrially scan data set. I
6:11evaluated each point cloud based off
6:13three criteria. Planar accuracy or the
6:16relative accuracy between flat features
6:19within a data set. Basically, what's the
6:22delta easting or northing between this
6:24wall and that wall or the delta
6:25elevation between the floor and the
6:27roof, but with 30 independent check
6:30shots that created a total of 44
6:32baselines per axis or 132 baselines per
6:36data set. This metric is basically what
6:39we're measuring when we use independent
6:41check shots via checkerboard targets
6:44scattered across a site, but since I'm
6:46using the one-dimensional method of
6:48accuracy determination, I'm able to
6:50remove virtually any influence from
6:53imperfect picking of the targets or any
6:56absolute shift that may be present in
6:58the data set I'm evaluating.
7:00Point cloud fidelity can be thought of
7:03as the scanner's ability to retain
7:05complex geometry in the point data. Most
7:09any scanner can do a decent job of
7:11capturing a flat wall, but piping,
7:13rafters, or really any detailed objects
7:16are a whole different scenario. This
7:18accuracy metric is all too often ignored
7:22in comparisons like these, yet it's
7:24vitally important when it comes to what
7:27we can actually use the point cloud for.
7:30And point cloud colorization will cover
7:32how well each scanner does at fusing the
7:36camera and point data together to paint
7:39the point cloud with correct RGB values.
7:42This is especially important for these
7:44slam scanners, as it's not uncommon that
7:47the features we want to extract don't
7:49have enough information from the
7:51geometry or intensity values alone to
7:55digitize everything we need to, and thus
7:58we are reliant on the RGB values of the
8:01point cloud to an extent.
8:03Speaking of intensity, we will also look
8:05at how good of a job each scanner does
8:08at assigning intensity values to each
8:10point in the cloud. What I'm trying to
8:12do with this comparison is essentially
8:14evaluate each data set based on the
8:17three pillars that affect the usability
8:20of a point cloud. To make this more
8:22digestible, I'll apply a point scoring
8:25system out of five based on how
8:27proficient each scanner is in its
8:29respective category. Now that we've
8:31covered what this competition is and how
8:33it was performed, let's cover what it's
8:35not. These weren't perfect lab-like
8:38conditions and as such, we can't expect
8:41optimal results or a representation of
8:44what these scanners are capable of at
8:47their very best. These were conditions
8:50that simulate a busy environment and the
8:53data was captured by representatives
8:55chosen by each company and thus some
8:57companies had more experienced users
9:00capturing the data.
9:02People make mistakes and if one were to
9:04rescan this site, it's quite possible
9:06that the results could be significantly
9:08different. Ideally, these scanners
9:10should be stable enough to preserve
9:12quality even if capture procedure wasn't
9:15ideal, especially considering how small
9:17the site was and the very high level of
9:20geometric diversity that surrounded it.
9:22And finally, it's important to remember
9:24this is just one data point in one
9:26specific environment and just because a
9:29scanner didn't do great here, doesn't
9:31mean it can't produce good data
9:33elsewhere using altered tactics. While
9:36it's not a perfect test, it's miles
9:39better than no test at all. There was a
9:42total of 10 scanners that entered the
9:44competition, nine SLAM scanners and
9:46Leica's new RTC700 that will be used as
9:49the source of geometric truth which had
9:52total cloud-to-cloud and bundle errors
9:55of 3 mm and that we will use to compare
9:58the other scanners against. Of those
10:00nine SLAM scanners, we'll separate them
10:01into two categories, the A-tier that
10:04uses the XT line of LiDAR pucks from
10:07Hesai. We have five units including the
10:09NavVis VLX 3 and MLX. I have full length
10:13videos on both of these, so I'll spare
10:15their introduction. We have the Green
10:17Valley LiGrip O2 Pro and the Fimes Slam
10:20200, which are your typical rotating
10:23head handheld XT based slam scanners,
10:26and the FJD T1 that was just released at
10:29Intergeo the day before this test took
10:32place. The T1 is quite interesting in
10:34that it can take stationary tripod
10:36mounted scans and use the redundant scan
10:40data to mathematically improve accuracy.
10:43The B-tier scanners are the smaller,
10:45lighter, cheaper counterparts of the
10:46A-tier and ditch the lidar pucks and
10:49instead opt for a dome style lidar
10:51sensor. In this category, we have the
10:54CHC Nav RS7 that uses Hesai's newer
10:58JT64P
10:59lidar sensor. We have the Manifold
11:02Pocket 2 Pro that Faro just white
11:04labeled as the Rov I1 that uses the
11:07Livox Mid-360, and we have the FJD X1
11:11and Navis CLX that were both just
11:13released right before this competition.
11:16The CLX is unique in that it's the only
11:18B-tier scanner that utilizes two Livox
11:21Mid-360s.
11:23As previously mentioned, the planar
11:25accuracy of the point clouds was
11:27measured using the one-dimensional
11:29method of point cloud accuracy
11:30determination. 30 independent check
11:33shots were extracted from the Leica
11:34RTC300 data set in between and away from
11:38the provided total station control
11:40points using n4ce's least square plane
11:43estimation tool that cuts through any
11:46noise in the data. These were the
11:47reference points, and the same procedure
11:49was done for each and every data set for
11:52the comparison points and that data was
11:55fed into my one-dimensional QC
11:57worksheet, which I provide for free via
11:59the video that references it, and it
12:02automatically runs a barrage of
12:03statistical analysis to thoroughly
12:06evaluate the accuracy of each data set.
12:09Starting off the A-tier with the Slam
12:11200 from Fima, it actually did a really
12:13good job with max residuals in each of
12:16the northing, easting, and vertical axes
12:18of 11, 8, and 10 mm, respectively, and
12:22standard deviations of just 2 and 3 mm.
12:26The Slam 200 kicks off this competition
12:28with a strong start. Next, we have the
12:30FJDTI 1 that struggled in this test. The
12:34T1 saw max residuals of 43, 68, and 29
12:39mm with standard deviations of 11, 17,
12:43and 8 mm. When we look at the histogram
12:46of the residuals, there appears to be a
12:48scaling issue where the T1 data set was
12:50too small compared to the control data
12:53set for whatever reason. I should also
12:55mention that I double-checked every
12:57large residual to confirm no errors were
13:00made during check shot extraction, even
13:02though Enforce is least squares plane
13:04fitting tool makes it pretty hard for me
13:06to make a mistake, thankfully.
13:08The LiGrip O2 Pro from GreenValley did a
13:11surprisingly good job with max residuals
13:14of 6 and 10 mm and standard deviations
13:18of those residuals of 1 and 3 mm. Before
13:22this test, I didn't really know too much
13:24about GreenValley, and as we dive deeper
13:26into the results, we'll see this is not
13:28a scanner to be ignored. Wrapping up the
13:31A-tier scanners, we have the MLX and VLX
13:333 from NavVis. The MLX had max residuals
13:37of 7, 8, and 11 mm with standard
13:40deviations of 2 mm in all three axes.
13:43The VLX 3 had max residuals of 8 and 10
13:46mm with standard deviations of 2 and 3
13:49mm, which means the LiGrip beat NavVis
13:52in this category and earned itself five
13:55points. I'll award four points to both
13:57the MLX and VLX 3, three points to the
14:00Slam 200 from Fima, and one point to the
14:03T1. Moving on to the B-tier of scanners,
14:06we'll start with the RS7 from CHC Nav. I
14:09had high hopes for this scanner, being
14:11that it veered away from the convention
14:13of using a Livox Mid-360 lidar sensor
14:16and opted for the newer Hesai GT64P.
14:19I got a chance to sit down with the
14:21engineers that designed the CLX while I
14:24was in Munich, and they did mention that
14:26they tried all possible lidar sensors
14:28that made sense in the CLX, and they
14:31kept coming back to the Livox Mid-360
14:34as being the option that provided the
14:37highest quality results. And considering
14:39the CLX certainly isn't the cheapest
14:41option in this year, and the engineers
14:43at NavVis certainly know what they're
14:45doing, I can't imagine they would have
14:47passed on the GT64 if it was in fact a
14:51higher quality sensor. After I saw the
14:54data set captured from the RS7, it more
14:56or less confirmed my revised
14:58expectations after my discussions with
15:01the NavVis team. It produced max
15:03residuals of 33 and 92 mm with standard
15:07deviations of 9, 12, and 23 mm. Like the
15:11residual histograms of the T 1, I saw
15:13some very obvious scaling issues. I
15:16brought my results back to CHC Nav, and
15:19they said the RS7 they used in this test
15:22was dropped earlier that day at the
15:24conference, which caused some hardware
15:26issues, and the calibration file of the
15:28scanner was no longer up to date. They
15:31tried to reprocess the data with a new
15:33calibration file, but there wasn't
15:35really any significant improvement when
15:37I QC'd the second data set. So, while
15:40the results in this specific planar
15:42accuracy test may not be indicative of
15:44what the RS7 is capable of, they were
15:47confident that the colorization and
15:49geometry preservation should not be
15:51affected by the scaling issue, and thus
15:53the other tests should represent the
15:55true capabilities of the scanner. Next
15:57up, we have the X1 by FJD that had max
16:01residuals of 9, 27, and 64 mm with
16:05standard deviations of 3, 7, and 15 mm.
16:09The Pocket 2 Pro from Manifold had max
16:11residuals of 23, 22, and 60 mm with
16:15standard deviations of 6 and 22 mm. Not
16:19very impressive, but keep in mind we
16:21must hedge expectations with B-tier
16:23scanners. There is a big difference
16:26between SLAM scanners using the Hesai XT
16:29line of lidar sensors and these
16:31cost-efficient dome lidar sensors.
16:34Last, but far from least, we have the
16:36CLX with max residuals of 12, 28, and 27
16:41mm with standard deviations of 3 and 7
16:44mm. Even the CLX had some pretty large
16:47residuals given the size of the site,
16:50but just like the RS7, NavVis mentioned
16:52after testing they found issues with the
16:55specific CLX that performed the scan. It
16:58took weeks, but it seems like they were
17:00able to fix the issue in post-processing
17:03for the most part. So, NavVis took the
17:05number one spot in the B-tier and earned
17:08itself five points for the CLX. The X1
17:10gets four points for its second place
17:12finish, and the Pocket 2 Pro gets three
17:15points with the RS7 coming up in last,
17:18and we'll give it one point for its
17:20relatively poor performance. To assess
17:22the ability of each scanner to retain
17:25complex geometry, we're going to pick
17:27five unique areas and section the data
17:30to determine how well each point cloud
17:33mirrors the actual geometry of the
17:35object it represents. With SLAM
17:38scanners, thin walls scanned from both
17:40sides often get smushed together during
17:43post-processing when the software can't
17:46distinguish between noise and two
17:48distinct planar surfaces. The first area
17:51we'll explore was an 18-mm wide wall
17:54panel attached to a wooden I-beam wall
17:57stud. We'll start with the A-tier
17:58scanners, and a reminder that we'll be
18:00using the RTC700 data as the frame of
18:04reference colored here in black. The
18:06SLAM 200 from Fyma didn't really merge
18:09both sides of the wall panel into a
18:10single flat plane, but there's so much
18:12noise that both sides look like one
18:15thick noisy plane, and the I-beam lost
18:17most of and shape. The FJD T1
18:21overcompensated the other way around and
18:23did turn both sides of the panel into a
18:25single thin flat plane. The I-beam was
18:27rounded out quite a bit, but not
18:29excessively by Slam standards. The Lie
18:32Grip 02 from Green Valley did the same
18:34thing more or less, but didn't turn the
18:36wall into a nice singular flat plane. It
18:39kind of zigzagged between the two faces
18:41of the wall and did preserve the I-beam
18:44better than the other two. The MLX was
18:46able to retain both sides of the panel
18:49as unique planes, although it enlarged
18:51that 18 mm wall to about 29 mm in places
18:56or an error of around 60% and the I-beam
18:59looks better than any previous data set.
19:02The VLX 3 was about the same with a
19:04marginal improvement and that will be a
19:06common theme going forward. It only
19:08enlarged the wall to 26 mm and it was an
19:11even expansion pushing north and south.
19:14I'll give four points to the VLX, three
19:16to the MLX, two to the Lie Grip, one to
19:19the T1 and zero to the Slam 200. For the
19:23B tier, the RS7 did a poor job in every
19:26aspect. It turned the I-beam into two
19:28triangles connected by a thin plane. It
19:31didn't separate the sides of each wall
19:33and overall produced an ugly and
19:35inaccurate result. The X1 on the other
19:38hand wasn't too bad, quite similar to
19:40what the T1 did which isn't surprising
19:42as they're both FJ D scanners and I
19:45would assume they share a lot of
19:46similarity in their Slam post-processing
19:49and filtering algorithms. The Pocket 2
19:52was pretty noisy on the I-beam. It was
19:54able to retain some very sparse data on
19:56that exterior wall plane and it only
19:59enlarged it by 1 mm. If it just did a
20:01better job on the I-beam, I'd be quite
20:03happy with this data from a B tier
20:05scanner. And the CLX actually did a
20:07great job for its class. It kept both
20:10sides of the wall with good density and
20:12enlargement of 6 mm and the I-beam looks
20:15good. It didn't preserve geometry as
20:18well as the VLX or MLX, but it's not a
20:21night and day difference mostly because
20:24the person scanning the site was able to
20:26get in nice and close to that beam. As
20:28we'll see in a minute, things change
20:30quite a lot when scanning geometry at a
20:32distance. Five points for the CLX, four
20:35for the Pocket 2, two points for the X1,
20:38and zero for the RS7. Next up, we have
20:40some small diameter copper water lines
20:43of various sizes scanned at close range.
20:46It's important to keep expectations in
20:48check with this test, and to be honest,
20:50I was kind of surprised NavVis even put
20:52these pipes in their testing area as
20:54anyone that's used SLAM knows how tough
20:57it is to get usable data on small
21:00diameter piping. The RTC even struggled
21:03a bit here, but was able to get workable
21:06data on even the smallest of pipes.
21:08Starting with the SLAM 200 from Fyma, it
21:11couldn't pick them up at all. We see a
21:13bit of extra noise where they should be,
21:15but that's about it. Even if we knew or
21:18estimated the pipe diameter,
21:19approximating them in the correct
21:21location would be inaccurate using this
21:24point cloud. This is essentially useless
21:27data. The T1 defined them significantly
21:30better. If their size was known, a
21:32skilled modeler could work with the data
21:34set as is, although accuracy would be
21:36affected. The LiGrip O2 was just about
21:39the same. And the MLX and VLX over
21:43smooth them and sucked them into the
21:45wall. If I knew those pipes were flat
21:47against the wall, I could have at least
21:49approximated their diameter by measuring
21:51how far the furthest grouping of points
21:53protruded from the plane of the wall and
21:56worked with the T1 or LiGrip data that
21:59way.
21:59That method would have given
22:01significantly incorrect diameters with
22:03the VLX or MLX data sets. Three points
22:07for the T1 and LiGrip, two for the MLX
22:09and VLX, and zero for the SLAM 200.
22:12Diving into the B tier, the Pocket 2 was
22:15almost as bad as the SLAM 200, and the
22:18lower we push the section away from the
22:20elevation where the scanner was held,
22:23the more the pipes melt into the wall.
22:25The RS7 failed miserably, nothing but
22:28noise. The X1 wasn't bad for a B-tier
22:31scanner. I could do something with this
22:33data. It wouldn't be great, but it'd be
22:35something. And I'd say the CLX was a
22:37touch better. It over-smoothed the pipes
22:39like its bigger brothers did, but
22:41there's something there. Again, keep
22:43expectations realistic. If you're hired
22:46for an MEP scan and they want half to
22:491-in piping and you show up with a SLAM
22:52scanner, you might want to rethink your
22:54tactics. Two points for the CLX and X1,
22:57zero for the other two. Next, we have
23:00some black boxes with sharp corners and
23:02the corrugated steel siding of the the
23:04C-can behind them. The SLAM 200 actually
23:07did a decent job here. The data is
23:09noisy, but the geometry is there. We can
23:12deal with noise, but we can't work with
23:14missing geometry. The T1 was about the
23:17same on the boxes, but rounded the
23:19corrugated siding a bit more than I'd
23:21like to see. This is a common theme with
23:23FJD's post-processing or filtering
23:26algorithm, or perhaps it was just the
23:28settings chosen while processing these
23:30specific data sets. The LiGrip was a
23:33touch behind the other two, wavy sides
23:35and some over-rounded corners, but the
23:37C-can looked good. The MLX and VLX were
23:40the obvious winners here. The VLX
23:42actually had a touch more noise on the
23:44boxes, but nothing I'd worry about. Five
23:47points for the MLX and VLX, four for the
23:49SLAM 200, three for the T1, and two for
23:52the LiGrip. As for the B-tier, the
23:55Pocket 2 did a decent job on the boxes,
23:57but not so great on the corrugated
23:59siding of the C-can. The RS7 was by far
24:03the worst of any data sets in this area,
24:06and on some sections of the C-can, it
24:08completely erased the corrugation. It's
24:10almost impressive how bad this scanner
24:12is doing, to be honest. The X1 did a
24:15great job for the most part. Some of the
24:17C-can was over-smoothed, but overall, it
24:20did better than the Pocket 2. And the
24:22CLX was a little rough and noisy, but
24:25I'd still give it the win here. Four
24:27points for the CLX, three for the X1 and
24:29Pocket 2, and zero for the R7. Let's
24:32jump up into the roof and see how these
24:34scanners fare at a further distance. We
24:36have some flat trim, corrugated roofing,
24:39and a skylight that were all about 15 m
24:41above the level the scanners would have
24:43been held at. The Slam 200 did okay, but
24:46the density was so poor, it made it hard
24:49to properly evaluate the data. I'm not
24:51exactly sure what caused this, but where
24:54there were points, I wasn't necessarily
24:55unhappy. Like we saw in previous
24:58examples, the T1 smoothing algorithm
25:00overdid it again. Great point density,
25:03but everything is a bit over rounded.
25:05The LiGrip did much better, not perfect,
25:08but so far it's leading the pack. The
25:11MLX did a pretty good job, but it also
25:14over smoothed the geometry compared to
25:16the LiGrip. The VLX preserved the detail
25:19in the corrugation a bit better than the
25:21MLX and LiGrip did. I'm guessing the
25:24chest-mounted LiDAR puck that directly
25:26scans overhead objects provided enough
25:29points to NavVis' post-processing
25:31algorithms where it didn't need to
25:34estimate the surface of the roof as much
25:36and could instead rely on actual point
25:39measurements. Five points for the VLX,
25:41four for the MLX and LiGrip, three for
25:44the T1, and one point for the Slam 200
25:47because the data was just too sparse,
25:50even though what was there looked kind
25:52of decent. None of the B-tier scanners
25:54did a good job on the roof panels. The
25:56Pocket 2 flattened the roof out and had
25:58a lot of noise. The RS7 was just
26:01terrible, noisy and flat. The X1 looks
26:04like it could have done a decent job,
26:06but there were some weird duplications.
26:09The CLX tried, but it's really tough to
26:12get decent data from a Livox Mid-360 at
26:1515 m away. The spot size is just too big
26:19and the angular accuracy is just too
26:21low. These are short-range LiDAR
26:23sensors. If you want to retain any kind
26:26of detail, 10 m is about my I with my P2
26:30when scanning detail and usually I'm
26:32aiming for about half of that. I'll give
26:34one point to the CLX and X1. Last but
26:38not least, we have some medium diameter
26:40drainage pipes. Before we get to the
26:42SLAM scanners, can we just take a moment
26:45to appreciate the RTC 700 data here?
26:49Leica has led the pack for years with
26:52the RTC 600 and the 700 made some big
26:55improvements to a tried and true
26:57scanner. Perhaps a dedicated RTC 700
27:00review video is in order in the not too
27:02distant future. The SLAM 200 did half
27:05decent on the larger two sets of pipes,
27:07but not so much on the smaller two. It
27:10would be tough to trust this scanner on
27:12piping under 100 mm based on what I'm
27:15seeing from this data set. The T1 did
27:18okay on the larger three sets, but
27:20shrunk each pipe. The LiGrip did a
27:23pretty good job on this one. It
27:24preserved size better than the T1 and I
27:27could do a decent job of digitizing even
27:29the smaller pipes. The MLX and VLX did
27:33well, but not amazing. They shrunk and
27:35distorted the smaller pipes, although
27:37the larger two were pretty good. I'd put
27:39them just behind the LiGrip just because
27:42it did preserve size better. Four points
27:44for the LiGrip, three for the VLX and
27:46MLX, two for the T1 and SLAM 200. The
27:50Pocket 2 is pretty noisy here. There was
27:52enough data to kind of work with it to
27:54an extent, but accuracy isn't going to
27:57be great. The RS7 did some weird things
28:00here and connected lines between two of
28:02the individual pipes. I'm guessing the
28:04software tried to fit a plane between
28:06the tangents of the smaller pipes. The
28:09X1 actually wasn't too bad. We had some
28:11overspray, but the geometry is there for
28:13the most part. The CLX was a little
28:15disappointing here. It squeezed the two
28:18smaller pipes into planes, which was a
28:20little odd. We'll do four points for the
28:22X1, three for the Pocket 2 and CLX, and
28:25two for the RS7. All right, let's tally
28:28it up. We have seven points for the SLAM
28:30200, 12 for the T1, 15 for the Li Grip,
28:3417 points for the MLX, and 19 points for
28:37the VLX-3.
28:39For the B-tier, we have the RS7 coming
28:41in hot at two points, and the Pocket 2
28:44has 10. That's 12 for the X1 and 15 for
28:47the CLX. To evaluate the colorization
28:50and intensity values of these data sets,
28:52we'll jump between common areas, and
28:54then fortunately, this one will have
28:56quite a bit of subjectivity. Remember,
28:58you can download and view all of these
29:00data sets via a Stitch 3D link in the
29:03description below. So, do that and make
29:06up your own mind. The colorization of
29:08the Slam 200 was fairly poor top to
29:11bottom. The camera was under exposed.
29:13There's a lot of smearing and artifacts
29:15where the images didn't stitch together
29:17too well. The intensity mapping was just
29:19okay. Not bad, not good. It's just low
29:22in detail compared to some of the other
29:24data sets that we'll get to in a second.
29:26The T1 was significantly better where
29:29the images stitched together well, but
29:31there was a ton of artifacting where
29:32they didn't. There was a bit more detail
29:34in the intensity mapping though. The Li
29:37Grip had good colorization. The white
29:39balance looks off, but overall, it's
29:41still good. It had a very narrow breadth
29:44of intensity values. I tried to tweak
29:46the visualization based on the intensity
29:48values I had to work with, but there
29:50wasn't a lot I could do to salvage this.
29:53A real shame because intensity values
29:55are quite important, and it did do a
29:57good job as far as the RGB values went.
29:59The MLX and VLX were about the same as
30:02each other and what you'd expect when it
30:04comes to NavVis. They have and will
30:06continue to set the bar when it comes to
30:09colorization. No one really comes close
30:11to them in this respect. Five points for
30:13the VLX and MLX, three for the T1 and Li
30:16Grip, and one for the Slam 200. The
30:19Pocket 2 had fairly impressive
30:21colorization for a B-tier Slam scanner,
30:23and after I adjusted the intensity
30:25visualization, it wasn't too bad there
30:28either. The RS7 was fairly
30:30disappointing. Smear colors, weak
30:33intensity mapping, not a great showing
30:35from this scanner throughout the test.
30:37The X1 wasn't quite as good as the
30:39Pocket 2 for colorization. It had some
30:41pretty hefty artifacts in certain
30:43places, but other places weren't that
30:46bad at all. The intensity mapping was
30:48the best we've seen from the B-tier so
30:50far though. And the CLX did what I
30:53expected any scanner from NavVis to do,
30:55produce a visually aesthetic point cloud
30:58with great RGB accuracy. The intensity
31:02mapping wasn't really any better than
31:03the X1, but I've no complaints here
31:06either. Five points for the CLX, three
31:08points for the X1 and Pocket 2, and one
31:11for the RS7. All right, let's see where
31:14we stand. I'm going to multiply the
31:16planar accuracy test results by a factor
31:18of five because while point cloud
31:20fidelity is critically important,
31:22without the building block of a good
31:24relative planar accuracy, it doesn't
31:26mean much. And there were five test
31:28areas and thus five opportunities to
31:30score for point cloud fidelity compared
31:33to one for planar accuracy. Colorization
31:35and intensity mapping will be multiplied
31:37by a factor of two because while it does
31:39matter, it's not as critical as the
31:41other two metrics of comparison.
31:43Starting with the A-tier in fourth and
31:46fifth place, we have the SLAM 200 from
31:48FARO and the T1 from FJD. Now, if I had
31:52to pick one of these two scanners to
31:54keep, I'm picking the T1 any day of the
31:56week. When I spoke to representatives
31:59from FARO, it was their opinion that
32:01most of the issues we saw in their data
32:03came from the methodology of capture the
32:06field tech used. Whether or not I agree
32:09with that may not be important and you
32:10can all make up your own mind there, but
32:12I'll say this. I can and have
32:15deliberately followed atrocious scan
32:17tactics with an FJD S2 and I don't get
32:20data that looks anything like what we
32:22saw from the SLAM 200. As far as the T1
32:26goes, it lost a ton of points in the
32:28planar accuracy test. They certainly
32:31need to sort out what happened there and
32:32fix that, but when they do, this scanner
32:35has great potential. And like I said,
32:37the T1 is brand spanking new. I know a
32:40thing or two about FJD and they are very
32:42good at improving their hardware via
32:45firmware updates. So, I'm very
32:46interested to see what comes of the T1,
32:49especially given the fact that it's
32:51capable of taking static tripod mounted
32:54scans. In third place, we have the
32:56LiGrip O2 Pro by GreenValley. I was very
32:59pleasantly surprised with this scanner.
33:02I didn't know much about GreenValley
33:04going into this competition, but I'll be
33:06keeping a closer eye on them. And in
33:08second place, we have the MLX, which
33:10means the NavVis VLX 3 takes first place
33:13in the A tier. Is anyone really
33:15surprised by this? Probably not. These
33:17tests gauge quality of data set, not
33:20portability, cost, flexibility, etc.
33:23NavVis has arguably always had SLAM
33:25scanners that produce the highest
33:27quality data sets with a few exceptions.
33:29Does that mean if I had to buy just one
33:31of these, it would be the VLX 3?
33:34Maybe, maybe not. Like I just alluded
33:36to, there are many more factors one must
33:39consider, but it's also pretty hard to
33:41go wrong with a VLX 3 or MLX, depending
33:44on what you're looking to do with
33:45either. Either way, congrats to NavVis
33:48and GreenValley on the podium finishes.
33:50Now, for our cheaper and smaller
33:52category, in a distant fourth place, we
33:54have the RS7 by CHC Nav.
33:57Like I said, I had really high hopes I
33:59was going to love the RS7 and that
34:01obviously wasn't the case. CHC Nav is a
34:04great company and one I worked with in
34:06the past on the AED 9 and iBase review
34:08video, but I have zero loyalty to any
34:11manufacturer and unfettered loyalty to
34:14the surveying industry. I'm sure CHC Nav
34:16won't like what they saw here today in
34:18terms of my analysis, but I hope they
34:20can go back to the drawing board and
34:22sort out the issues we saw from this
34:24data set and come back in 2027 to kick
34:27some ass. In third place, we have the
34:30Manifold Pocket 2 Pro or the Faro Rho Vi
34:331, depending on the badge that's on it
34:35that day. This isn't a bad scanner at
34:37all. I think anyone would be happy with
34:39it as long as they have realistic
34:42expectations of a live aux mid 360
34:44scanner. The X1 from FJD earned a second
34:47place finish for the B tier. It was
34:49decently accurate, preserved geometry
34:52quite well, and appears to be a solid
34:54improvement to their P2. It certainly
34:56has improved ergonomics with an
34:58integrated screen and larger internal
35:00cameras. Congrats to FJD on the
35:02runner-up spot. And again, this is a
35:04brand new scanner and will only get
35:06better as time goes on. This last one
35:09didn't surprise me at all. The CLX takes
35:11the first place spot for the B tier.
35:13It's the best dome scanner on the market
35:15hands down, at least that I've seen and
35:18I've seen and tested a lot. It's still
35:20limited by what a live aux mid 360 can
35:22do. So, if you're shopping for one, keep
35:25in mind you're not going to be able to
35:27retain nearly as much detail at a
35:29distance or accuracy compared to the MLX
35:32or VLX. NavVis is a champ at
35:35post-processing and they have a solid
35:38SLAM algorithm, but they can't break the
35:40laws of physics. Well, that wraps up the
35:422026 third dimension SLAM off. If any
35:45manufacturers are interested in joining
35:47the 2027 SLAM off, please feel free to
35:50reach out. And I'd like to thank each
35:52and every company that participated.
35:54They all took a chance doing so as I was
35:56clear I was going to be honest about
35:58what I saw, and this was a competition
36:00after all. Not everyone can win. If
36:03you're wondering why AccuGrid's,
36:04OmniSLAM, Emesent, Share, etc. weren't
36:07in the competition, let's just say
36:09plenty of other invitations to join were
36:12offered while I was at Intergeo, but the
36:14list we ended up with were the ones that
36:17were confident enough in their products
36:19to put them on display for truly
36:21objective testing. And that kind of
36:24transparency really means something in
36:26an industry that's known for big claims,
36:29but always seems to be short on proof to
36:31back them up. We're constantly hearing
36:33some companies talk about 1 mm or 2 mm
36:37accuracy, but when given the chance to
36:39prove it in a video that will be watched
36:41by tens of thousands of people in the
36:43industry, why not take the opportunity
36:46to do so if you actually believe those
36:48claims? And even though Fyma or CHT Nav
36:51might not have had a great showing this
36:53time, this is just one data set on one
36:56day. Take it for what it's worth, and I
36:58hope to see them and the others back
37:01next year along with some of those
37:03companies that like to talk the talk,
37:05but whose legs get tired when it's time
37:07to walk the walk. I'm Shea Gleedle of
37:10The Third Dimension signing off until
37:12next time.

About this transcript

This page contains the full transcript of SLAM WARS '26 : 9 Handheld LiDAR Scanner Competition (NavVis vs CHCNav vs FJD vs GreenValley & more) by The 3rd Dimension, generated from the public captions YouTube serves with the video. The transcript has 6,290 words across 899 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.