What is SLAM? Simultaneous localization and mapping explained for surveyors

What is SLAM (simultaneous localization and mapping)?

Simultaneous localization and mapping is the computational problem of building a map of an unknown environment while tracking the sensor's position inside that map at the same moment. The two tasks depend on each other: you need a map to know where you are, and you need to know where you are to place new measurements on the map. SLAM solves both together.

The term comes from robotics research in the late 1980s and 1990s, where mobile robots had to navigate rooms without a floor plan. It is sometimes expanded as simultaneous location and mapping, and in British English it is written simultaneous localisation and mapping; all three describe the same idea. Today the same algorithms run in robot vacuum cleaners, autonomous vehicles, drones and, most relevant to us, handheld and backpack scanners used for surveying.

For a surveyor, the practical meaning is simple: a SLAM mapper lets you walk through a building, tunnel or forest and come out with a georeferenceable 3D point cloud, without setting up a tripod at each station. If you want the field side of this, our guide to SLAM scanning and mobile mapping covers walking routes and project planning; this article explains what happens inside the algorithm.

How does SLAM localization and mapping work step by step?

SLAM localization and mapping run as a continuous loop: predict the motion, measure the surroundings, match the measurement to the map, update the position and the map, and periodically correct the whole trajectory. A typical lidar-inertial SLAM pipeline looks like this:

1. Motion prediction. The IMU (accelerometers and gyroscopes) measures movement hundreds of times per second and predicts where the scanner has moved since the last lidar frame.
2. Scan capture. The rotating lidar head records a frame of range measurements, often hundreds of thousands of points per second, around the operator.
3. Feature extraction. The software picks stable features such as planes (walls, floors, ceilings), edges and corners, and discards moving objects where it can.
4. Scan matching. The new frame is aligned with the local map, using the IMU prediction as a starting point. This gives the refined position and orientation (the pose).
5. Map update. The aligned points are added to the map, which grows as you walk.
6. Loop closure. When the system recognises a place it has already mapped, it measures the mismatch between the two passes.
7. Global optimisation. A pose-graph optimisation spreads that correction back along the whole trajectory, so the map closes neatly instead of showing doubled walls.

Steps 1 to 5 are often called the front end and run in real time; steps 6 and 7 are the back end, which is why post-processed SLAM results are usually cleaner than the live preview.

What are the main types of SLAM?

SLAM systems are grouped by the sensor that observes the environment: lidar SLAM, visual SLAM, or a fusion of several sensors. Survey-grade mobile mapping almost always uses lidar with an IMU, because lidar measures distance directly and works in the dark.

3D SLAM is the category that matters for surveying, BIM and mining. Visual SLAM is common in phones, augmented-reality headsets and small robots, but it is rarely used where a client specifies point-cloud accuracy.

How accurate is SLAM, and what causes drift?

A good lidar SLAM system typically achieves 1 to 3 cm relative accuracy in well-structured environments, but its error grows with the distance walked unless loop closures or control points correct it. That growing error is called drift.

Drift comes from small errors in every scan match and in the IMU, which add up along the trajectory. It is worst in places where the geometry gives the algorithm little to lock onto:

• long, uniform corridors and tunnels where every frame looks the same along the axis;
• open car parks, fields and flat roofs with few vertical features;
• stairwells walked too quickly, with sharp turns and sudden changes in view;
• scenes full of moving people, vehicles or vegetation in strong wind.

In our engineers' experience, the single biggest accuracy gain comes from planning closed loops: start and finish at the same spot, and cross your own path every few minutes. Our handheld SLAM laser scanner is specified at a point-cloud thickness of 1 cm or better after post-processing and 2 cm or better in real time, with a range of 0.1 to 70 m. Treat such figures as relative precision under good conditions; absolute accuracy also depends on how the cloud is tied to control.

How do you georeference a SLAM point cloud?

A SLAM point cloud is created in its own local coordinate frame, so to put it on a project grid you tie it to surveyed control points. This also constrains drift on long routes.

1. Establish control targets or marks along the route with a GNSS RTK receiver outdoors or a total station indoors.
2. During the scan, pause on each control mark (or scan identifiable targets) so the software can tag them.
3. In processing, enter the control coordinates and run a constrained adjustment of the trajectory.
4. Check residuals on independent check points that were not used in the adjustment, and report them.

A practical rule we give customers is one control point at the start and end of every loop, plus one roughly every 50 to 100 m on long linear routes such as tunnels. For general registration and cleaning steps, see our article on point cloud processing.

What are the main simultaneous localization and mapping applications?

The main simultaneous localization and mapping applications are robot and vehicle navigation, and fast 3D capture of spaces where tripod scanning or GNSS is slow or impossible. In surveying and engineering the common jobs are:

• Building surveys and BIM: multi-storey interiors, plant rooms and risers captured in a single walk, then modelled.
• Underground mining and tunnels: drives, stopes and tunnel profiles where there is no GNSS.
• Forestry: tree height, stem diameter and stand density under canopy.
• Construction progress: weekly walk-throughs to compare built work with design.
• Stockpiles and sheds: volumes inside covered storage, linked to our guide on stockpile volume measurement.
• Heritage and facilities: complex buildings documented without hours of setups.

Outside surveying, SLAM drives warehouse robots, autonomous cars, delivery drones and augmented reality, but those consumer and automotive uses have different accuracy goals.

SLAM vs terrestrial laser scanning: which should you use?

Use SLAM when speed and coverage matter more than millimetres; use a tripod-based terrestrial laser scanner when you need the highest accuracy and density on each surface. Many projects combine both.

A common hybrid workflow is to scan critical areas, such as a façade or a mechanical room, with a high-speed terrestrial laser scanner and fill the rest of the building with SLAM, registering both to the same control. Our article on terrestrial laser scanning explains the static workflow in detail.

Common SLAM mistakes in the field and how to avoid them

Most poor SLAM results come from how the route was walked, not from the hardware. SLAM engineering assumes smooth motion and enough geometry in view, so field habits matter.

• Walking too fast or turning sharply: keep a steady walking pace and turn slowly at doors and corners.
• Passing through doors without pausing: stop briefly in doorways so the scanner sees both rooms.
• No loops: design routes that close back on themselves; open-ended routes drift.
• Starting while moving: initialise the scanner while standing still for the time the manufacturer specifies.
• Ignoring control: without control points the cloud cannot be checked or placed on the project grid.
• Too-long sessions: split very large sites into overlapping missions rather than one huge walk.

If you are deciding whether SLAM fits your work, or want to compare it with other 3D laser scanners, browse our 3D laser scanner range or request a quote with your project type and site size, and our engineers will recommend a setup and control strategy.

Frequently asked questions