Indoor Space Mapping with Handheld LiDAR: A Practical Guide

Half the quality of your point cloud is decided before you press record. Not by the sensor specs, not by the SLAM algorithm — by how you walk through the building.

I've seen a 5cm-accuracy scanner produce a point cloud that looks like it went through a blender, just because the operator picked a bad initialization point in a featureless hallway. I've also seen the same hardware nail a 30-room office scan in 40 minutes because the operator planned the path, kept their speed steady, and closed their loops. The difference is technique, not equipment.

This guide covers the stuff that actually matters when you're holding a LiDAR scanner and walking through someone's building: how to plan the scan, how to walk, how to handle the environments that wreck SLAM algorithms, and what to do with the data afterward. I'll reference the M360 and M360-D (dual-echo) specs where they're relevant, but the techniques here apply to any handheld SLAM scanner.

Pre-Scan Planning

Walk the site before you power on the scanner. Five minutes of reconnaissance saves two hours of re-scanning, and it's the cheapest insurance you can buy.

Open every door you plan to scan through. Turn on all the lights. Move rolling chairs, trash cans, and anything that might shift between passes. If you're scanning a multi-room floor, open all connecting doors beforehand so you don't have to stop mid-scan to fumble with keys.

While you walk through, note the problem areas:

• Glass walls and curtain walls. These reflect LiDAR pulses in unpredictable ways and create phantom geometry in your point cloud. Know where they are before you start.
• Repetitive corridors. Long hallways with identical doors, lights, and wall features are where SLAM drift compounds. You need a strategy for these.
• Narrow stairwells. Tight vertical spaces force you to move slowly and hold the scanner in awkward positions.
• Mirrors and polished surfaces. Lobbies with decorative mirrors, elevators with polished metal doors — these create false returns.

Plan your scanning sequence and don't change it mid-scan. Start from the most complex part of the building (usually the core with stairwells and elevator lobbies) and work outward. The reason: if something goes wrong and you need to restart, you've already dealt with the hardest part and your muscle memory for the path is fresh.

Write the sequence on your phone or a notepad. Room 1 → Room 2 → corridor → Room 3 → back to Room 2 → exit. Something like that. You'd be surprised how quickly you lose track in a 40-room building.

Initialization — Where to Start

The SLAM algorithm needs a good initialization point to anchor the rest of the scan. Get this wrong and the entire point cloud drifts, sometimes progressively, sometimes in one sudden jump.

Initialize in a feature-rich area. A conference room with furniture is ideal. A table surface provides geometric complexity at scanner height, which gives the algorithm more constraint than a flat floor. The ceiling-wall-floor corner of any room works well too.

Avoid initializing within 1 meter of a wall. Close-range points have higher measurement noise, and the algorithm can get confused when the geometry is dominated by one big planar surface right next to the sensor.

Stay away from mirrors, glass, and anything highly reflective. If your starting room has a wall of windows, face the scanner toward the interior wall, not the glass.

Don't initialize near moving targets. Someone walking past, a door swinging shut, a ceiling fan spinning — any of these can corrupt the initial state. Wait for stillness.

Walking Speed and Technique

For general indoor spaces — offices, retail, residential — walk at 0.5 to 1.0 m/s. That's a deliberately slow pace, somewhere between a stroll and a normal walk. If you're moving faster than 1.0 m/s, the SLAM algorithm is spending more computational budget on motion compensation and less on loop closure. The point cloud gets noisier.

Drop to 0.5 m/s or slower in narrow corridors (under 1.5m wide), staircases, and tight spaces where geometric features change rapidly. On stairs, count your steps if you need to — it forces a slow, consistent rhythm.

Don't run. Don't stop suddenly. Don't do sharp 90° turns. All of these cause point cloud artifacts because the SLAM algorithm's motion model assumes smooth, continuous movement. Sudden changes break that assumption and the point cloud "stretches" at the discontinuity.

Hold the scanner stable. Natural arm swing while walking is fine — the IMU compensates for it. What you don't want to do is deliberately swing or rotate the scanner to "cover more area." That just injects angular noise and confuses the motion estimation.

For narrow doorways and corridors: hold the scanner against your chest, walk sideways, and let the LiDAR's vertical FOV do the work. The M360's 70° vertical FOV (-10° to +60°) means it captures both the floor and ceiling even at close range, so you don't need to tilt it to see both surfaces in a tight space.

The 5cm blind zone on the M360 matters here. Most handheld scanners have blind zones of 20cm or more, which means you lose geometry right around doorframes, table edges, and other close-range structures. With a 0.05m blind zone you preserve that near-field detail — wall thickness, doorframe geometry — which is exactly what architectural clients want to see.

Path Strategy — The Loop Closure Rule

Every scan should form a loop. You start at point A, walk the path, and return to point A. When the SLAM algorithm recognizes that you've returned to a previously scanned area, it corrects accumulated drift by aligning the new observations with the old ones. This is called loop closure, and it's the single most important concept in handheld SLAM scanning.

An O-shaped loop is better than retracing your steps. When you walk back and forth over the same path, the algorithm is constantly trying to resolve the overlap, which can introduce oscillation errors. A loop gives the algorithm one clean closure opportunity at the end.

For scans covering more than 100 meters of path, create intermediate loop closures every 100-200m. This prevents drift from compounding to the point where the final loop closure can't resolve it. Think of it as saving your work incrementally instead of hoping one big save at the end catches everything.

For multi-room buildings, use a room-by-room strategy with small loops inside a big loop. Enter Room 1, scan the perimeter, return to the doorway. Enter Room 2 from the same doorway, scan, return. The doorway becomes a persistent loop closure point each time you pass through it, letting the algorithm re-register against known geometry.

Always return to the last doorway before entering the next room. This is the rule that most beginners break, usually because they want to "save time" by cutting across. Don't cut across. The loop closure at the doorway is more valuable than the time you save.

Scan one floor at a time. Do not attempt a multi-floor scan in one continuous recording. The vertical transition introduces motion model errors that are difficult for most SLAM algorithms to handle. Scan floor by floor, then register the floors in post-processing.

Problem Environments and How to Handle Them

Not all buildings cooperate. Here are the environments that cause the most trouble and what to do about them.

Glass curtain walls and mirrors

Glass is the enemy of single-echo LiDAR. A pulse hits glass and most of the energy passes through — you get a weak, noisy return from the glass surface, or worse, a return from whatever is behind the glass. Mirrors are even worse because they reflect the pulse at an angle, producing phantom geometry that appears meters away from the actual mirror surface.

The M360-D dual-echo variant handles this significantly better. Dual echo captures both the first return (the glass surface) and the second return (the object behind the glass), so you can separate real geometry from phantom geometry in post-processing. For buildings with lots of glass — modern offices, atriums, storefronts — dual echo is the better choice.

For single-echo sensors, adjust your scanning angle. Instead of walking straight toward a glass wall, approach it at a 30-45° angle. This increases the chance of getting a usable first return from the glass surface. Also, note the glass locations on your floor plan and treat those areas with skepticism during point cloud cleanup.

Long repetitive corridors

Office buildings, hospitals, and schools love long corridors with identical doors, lights, and fire extinguishers every few meters. The problem is that the SLAM algorithm relies on distinctive features to track its position. When every 10-meter section looks identical, drift accumulates silently.

Add reference objects every 50m. Place a chair, a box, or anything with a distinctive shape in the corridor before scanning. These break up the monotony and give the algorithm anchor points. If you can't modify the environment, walk even slower (0.3-0.5 m/s) and plan your loop closures so you return to the corridor entrance frequently.

Epoxy-coated garage floors

Epoxy floors are common in parking garages and some industrial facilities. They're highly reflective at the LiDAR wavelength, which produces dense, noisy floor points that can overwhelm the algorithm and degrade point cloud quality.

Tilt the scanner slightly upward — maybe 10-15° — so the vertical FOV favors wall and ceiling geometry over the floor. You'll lose some floor coverage, but the floor is usually planar and can be reconstructed from fewer points anyway. The walls and ceiling carry the structural information that matters.

Staircases

Scan stairs from bottom to top, not top to bottom. Keep the LiDAR facing the inner handrail side so the sensor has continuous wall and railing geometry to track. Don't block the LiDAR's view with your body — hold the scanner out slightly from your chest on the handrail side.

The M360's 5cm blind zone helps here. At that range, you can capture the handrail geometry without losing it, which is a common problem with sensors that have 20cm+ blind zones.

Large open rooms and warehouses

Big empty rooms — gymnasiums, warehouses, atriums — feature vast expanses of flat floor and flat ceiling with very little geometric variety in between. The algorithm starves for features.

If the room has shelving, equipment, or any vertical structures, scan along the rows rather than through the empty aisles. Walking down a 2m-wide aisle between shelving units gives the algorithm wall-to-wall geometry at scanner height. Walking through the center of an empty warehouse gives it nothing but a floor and a ceiling.

For truly empty spaces, place temporary reference objects — boxes, ladders, chairs — along your path at regular intervals. Remove them afterward in post-processing. It's tedious, but it produces a usable point cloud where no features exist.

SLAM Algorithm Tips

Most handheld LiDAR systems use one of two open-source SLAM algorithms: FAST-LIO2 or Point-LIO. Both are designed for solid-state LiDAR sensors like the Livox pattern (which the M360 uses), and they handle the non-uniform point distribution pattern well.

FAST-LIO2 is the default choice for general indoor scanning. It runs well on modest hardware, works with the built-in IMU (the M360 has a 6-axis IMU: 3-axis accelerometer + 3-axis gyroscope), and handles moderate walking speeds without trouble. For offices, apartments, retail spaces, or any typical indoor environment, FAST-LIO2 gets the job done.

Point-LIO is better suited for corridor-heavy environments and slower, more deliberate scanning. It's more computationally intensive but produces tighter point cloud registration in feature-sparse areas. If your building is mostly long corridors (hospitals, schools, hotels), Point-LIO might give you better results at the cost of needing a more powerful compute platform.

Some systems offer real-time point cloud preview. If yours does, use it. Watch the point cloud as you walk and look for sudden drift, stretching artifacts, or misalignment at overlaps. Catching a problem during the scan is far better than discovering it during post-processing.

Post-Scan Workflow

Once you've walked the building and closed your loops, the raw data needs processing. Here's the typical workflow:

1. Export the point cloud from the SLAM system. Common formats are .las, .ply, and .e57. E57 is the standard for architectural deliverables. LAS is more common in survey and GIS workflows. PLY is universal and lightweight.
2. Remove noise and outliers. Most SLAM software has automated outlier removal tools. Run them, then visually inspect the result. Pay special attention to glass walls, mirrors, and highly reflective surfaces where you'll see ghost points floating in space. The dual-echo data from an M360-D makes this step easier because you can filter by first/last return and separate real surfaces from behind-glass artifacts.
3. Register multiple scans if needed. If you scanned the building in multiple sessions (different floors, disconnected areas, or because the battery died mid-scan), register the point clouds using common overlap areas. CloudCompare is free and handles this well — use its loop closure optimization on the registered point cloud to tighten up any residual drift.
4. Extract deliverables. What the client needs depends on the project:
   • Floor plans: Section cuts through the point cloud at 1.0-1.2m height, traced in CAD software.
   • Sections: Vertical cuts for elevation drawings.
   • 3D models: Mesh the point cloud in MeshLab, Blender, or commercial software like RealityCapture. Clean up the mesh, apply textures if you have photos, export as OBJ or FBX.

The M360's detection range of 0.05–50m (at 90% reflectivity) and 0.1–25m (at 10% reflectivity) means you're getting usable returns from everything in a typical indoor environment. The 50m upper limit is more than enough for even large warehouse bays. The 5cm minimum range means you're not losing the close-range geometry that matters for architectural detail.

Before Your Next Scan

A practical checklist. Not rules — just habits that make scanning less painful.

1. Walk the site before scanning. Note glass, mirrors, repetitive corridors, and narrow spaces.
2. Open all doors and turn on all lights.
3. Plan your path and write it down. Start complex, end simple.
4. Initialize in a feature-rich area, away from walls, glass, and moving objects.
5. Walk at 0.5–1.0 m/s. Slower in tight spaces.
6. Close every loop. Return to your starting point or the last doorway.
7. Scan one floor at a time.
8. Watch the real-time preview if available.
9. Export in the format your client needs. E57 for architecture, LAS for survey, PLY for general use.
10. Clean up glass and mirror artifacts before delivering.

The sensor does its job. Your job is to give the sensor a reasonable environment to work in. Walk slowly, plan your loops, handle the problem areas deliberately, and the point cloud takes care of itself.

Need help choosing the right LiDAR sensor for indoor mapping? Check out our M360 vs MID-360 comparison or contact our team to discuss your project requirements.

© 2026 SmartBotParts. All rights reserved.