I Bought a $5,600 SLAM Scanner and Still Can't Make It Pay for Itself

So back in November I picked up a SHARE S20. First one through the door outside of demo units. Rami at The Survey Skool -- yeah, that Rami Tamimi from YouTube -- we had been talking about the unit. He said Share had sent him one for testing and I had a use case so I bought one. I figured I'd either make something of it or break it trying. [Affiliate link — Wet Dog Drone Services may earn a commission if you sign up for the Survey School through this link, at no cost to you. We did not commission, pay for, or otherwise compensate Rami or the Survey School for the making of the linked video. Wet Dog Drone Services is an affiliate of the Survey School community.]

I broke nothing. The jury's still out on the second part.

Here's the deal after running this scanner for six months on real jobs: the S20 is genuinely good hardware with immature software, a processing pipeline held together by duct tape and a Python script that's on version 89, and a value proposition that's crystal clear -- if you're willing to wait for the market to show up.

Let me back up.

What the S20 Actually Is

Livox Mid-360 solid-state LiDAR. 200,000 points per second. 40-meter range for data you'd actually stake your reputation on, 70 meters if you're feeling optimistic. Two 1-inch CMOS cameras with mechanical shutters that shoot photos to colorize the point cloud. The whole thing weighs maybe two pounds and fits in one hand.

About $5,600 to $6,000 depending on tariffs and which accessories you bundle. The processing software -- Point Cloud Studio -- is currently free.

The mechanical shutter is the thing that matters. If you've done any photogrammetry, you know motion blur is the enemy. Electronic shutters do that rolling shutter thing that turns your nice crisp points into smeared garbage when you're walking at full stride. The S20 doesn't have that problem. Two 1-inch sensors, mechanical shutters, clean images. Done.

That's the hardware pitch. It's a good one.

The Scanning Workflow (Field Side)

You set the S20 on a flat surface to calibrate. Pick it up. Walk. That's it.

I scanned 120,000 square feet of interior space -- multiple floors of a gutted building -- in two days. For wide-open areas like parking garages, the workflow is stupid simple: middle of every other parking spot, systematic pattern, move on.

The data coming out is clean. I was genuinely shocked the first time I compared it against a Navis unit I'd demoed. Navis data is amazing -- voodoo blackmagic bullshit clean. But the S20 isn't far behind, and it costs a fraction. I compared it against a Leica BLK too and was less impressed with the Leica. Benchmarks can always be rigged to make the competition look bad, but the S20 earned its "front runner at its price point" label honestly.

Here's the thing that surprised me: the noise. Handheld SLAM scanners should be noisy. You're walking, the IMU is bouncing, the LiDAR is doing its best to track trajectory while your body introduces all kinds of micro-movements. But the point cloud that comes out of this thing is not that noisy. I wasn't expecting that.

Where the Software Fights You

Point Cloud Studio processes the data after you bring it back from the field. And this is where the "first-generation product" label kicks in.

Early versions of the software would silently save a new point cloud file every time you cleaned up data. Clip some noise, it auto-saves. Clip some more, auto-saves again. You'd walk away from a one-hour cleanup session and find your hard drive 200 gigs heavier for no reason. They've since added a separate save button to stop that nonsense.

Colorization on flat surfaces was another one. I had a scan of an underground vault area in downtown Denver, where the wall colors came out wonky -- different shades of "not what the wall actually looks like" across what should have been a uniform surface. I took screenshots, sent them to Share, and they confirmed it was a known issue that's since been fixed. But I encountered it on a live job with a real client waiting for data. That's the risk with early-stage software.

The photo interval defaults to 2 seconds. Sounded fine. I walked a site at normal speed, brought the data into Pix4D for processing, and got failures. Turns out 2 seconds at walking speed doesn't give the software enough overlap between images to reconstruct geometry properly. User error -- I should have known better. But the software doesn't warn you. It just fails.

Control Points Are Not What You Think They Are

This one cost me twenty minutes of confused conversation at dinner. When I say "control point," I mean a georeferenced position on the planet that I've shot with a GNSS receiver and know to centimeter-level accuracy. That's what anyone in this space means.

Share's idea of "control points" is four big squares -- 18x18 or 24x24 -- that you lay on the ground side by side to make one big square. You set the S20 down on one corner to start your scan. You walk. You come back and set it down on the opposite corner to close the loop. Those squares exist to help you merge two scans together.

That's it. Not georeferencing. Scan merging.

I had a long conversation with Mark, one of their engineers, to hash this out. The scans do merge together well using those targets. But if you want real-world coordinates on your data, you need to shoot actual control points with a GNSS receiver and tie them in separately. Or you merge your SLAM data with a drone point cloud that already has coordinates -- which is what I do.

Merging S20 Data With Drone Data

This is where the scanner earns its keep in my workflow.

When I fly a drone over a construction site, I get great data on top of everything. But I always get holes. Undersides of overhangs. Interiors of walls right next to buildings. Areas blocked by cranes or equipment. These gaps are a given in photogrammetry. You can't map what you can't see from the air.

The S20 fills those gaps. I walk the site with the scanner, capture all the interior spaces and obscured areas, clean up the SLAM output, and merge it with the drone point cloud in CloudCompare using ICP -- Iterative Closest Point registration.

Workflow for a recent job: I flew a site, maybe 10 acres inside the fence line, 17.5 total including surrounding areas. I shot four to five targets for alignment in Pix4D, but I had 23 permanent control points across the site -- manhole covers, water valves, storm sewer grates. The S20 data gets cleaned, imported into CloudCompare, merged with the drone data using shared control points as tie-ins. Then the whole package goes to the superintendent and the engineer.

The result is complete coverage. Aerial from the drone, ground-level and interior from the S20. Fills every hole.

Gaussian Splatting: The Real Reason I'm Holding Onto This Thing

Here's what got me. I show a GC a traditional point cloud -- 50 million points, dense, technically precise. He stares at it for thirty seconds and says, "So where's the building?"

I switch to a Gaussian splat. He "walks" through the site on his phone. Five minutes later he's approving the budget.

That's it. That's the difference.

Gaussian splats are photorealistic 3D visualizations that run in a browser. They're not meshes. They're not point clouds. They look like someone built a video game level of your job site. For a decision-maker who needs to understand spatial relationships without taking a graduate course in point cloud interpretation, splats are "10 times better."

The catch: processing. I ran a 5-6 minute walk through a church interior through Gaussian splatting. Took 33.5 hours at 96% utilization on an RTX 5080 with 96GB of RAM. That's overnight processing for about 10,000 square feet. It doesn't scale yet.

And the pipeline isn't turnkey. To get professional-quality splats from S20 data, you export from Point Cloud Studio, convert to COLMAP format (there's a community Python script for this -- currently on version 89 for a reason), then feed it into Reality Capture 2 or Postshot for camera alignment and training. It works. It's just not one-click.

For a recent client meeting, I showed both the raw point cloud and the Gaussian splat. The splat is what closed the deal. Bottom line money -- if the client can't see it, they won't buy it.

What I'd Tell Someone Buying the S20 Today

Don't buy it expecting Gaussian splatting to be turnkey. It's not. Don't buy it if your primary deliverable is 2D plans -- the S20's value is in 3D. Don't buy it for open terrain or large-scale topography -- a drone with RTK rover does that faster and cheaper.

Buy it if you're chasing the remodel/TI market. Pre-bid interior scanning for GCs who are tired of estimating from wrong drawings. That beachhead is real.

Buy it if you already run drone scanning and need to fill coverage gaps. S20 plus drone plus CloudCompare is a complete solution for complex sites.

And wait for the K2 or S30 if you need more range. The S100 hits 300 meters with a mechanical LiDAR at 1.15 million points per second. The S20 does 40 meters at 200,000. That's a 7.5x range difference and a 5.7x point density difference. It matters if you're scanning building facades or tall structures.

The Honest Bottom Line

I've scanned 120,000 square feet in two days. I've merged SLAM data with drone data on active construction sites. I've watched a splat walkthrough close a meeting that a point cloud couldn't. And I still haven't made this scanner pay for itself yet. The math doesn't work at current volume.

But I'm not selling it. I've been tempted -- the K2 announcement had me looking at upgrade paths, and if I'm being honest the thought has crossed my mind more than once. But every time I pull splat data back up on a screen and watch someone immediately understand a job site, I know the market is coming.

The technology works. The hardware is solid. The software is catching up. What's missing is the ecosystem -- turnkey processing, standardized deliverables, and the GC who automatically specifies SLAM scanning in their bid package.

That's coming. And when it does, the people who learned this tool early are going to own that market.

If you want to go deeper on the field workflows — the actual step-by-step process of scanning, processing, and merging SLAM data with drone data — the SHARE Scanner Field Manual is a community-sourced practical guide that covers the full workflow. It's a real resource built by people using this gear on real jobs, not a marketing brochure.

Thumbs up buttercup. Go scan something.

---