10.8 Accuracy Reference
What the published specifications actually mean, the conditions that produce them, how to choose a control method for each AEC project type, and how to verify and document the accuracy you delivered.
What the Published Specifications Actually Mean
The accuracy figures published by XGRIDS are achievable results under defined conditions. They are not guaranteed outcomes from any scan regardless of technique, environment, or control strategy. Every figure below carries the condition that produces it, because a number quoted without its condition is not the number XGRIDS published.
RMSE figures also assume proper field procedure and unbiased error distribution. Real-world results can be affected by atmospheric conditions, antenna geometry, multipath from nearby structures, satellite count, baseline length to the correction source, and CRS or geoid configuration errors. Any of these can introduce systematic error that degrades outcomes below the published specification regardless of Fixed status or GCP count.
L2 Pro
K2
The K2 control point interval is half the L2 Pro interval. Both devices reach 3 cm absolute accuracy, but the K2 needs twice the control density to get there. On a site you have previously scanned with an L2 Pro, a K2 rescan using the same control network will be under-constrained. Plan the network against the device that will actually be carried, and if both devices work the same site, build the network to the K2 spacing so either can use it.
K1 (Legacy)
K1 is no longer in production. Specifications retained for existing K1 owners.
PortalCam
The PortalCam has no accuracy specification, and that is a scoping fact rather than a gap. It produces 3D Gaussian Splat output only, so there is no georeferenced coordinate system for an absolute figure to describe. It does use control points, but they connect segments during Map Fusion rather than tie the model to a survey coordinate system. The LCC Studio measurement tool returns approximate distances that are not survey-grade. The two figures below describe internal geometric consistency for visualization work. Do not quote them in a proposal, a scope of work, or a tolerance clause. If a client needs a number they can hold you to, the job needs an L2 Pro or a K2.
Relative vs. Absolute Accuracy
Relative Accuracy
Relative accuracy is the internal consistency of the point cloud: whether a wall is planar, whether measured dimensions repeat reliably, whether the geometry holds across the full scan without distortion. A scan with good relative accuracy looks correct and measures correctly within itself, even if it is not registered to any real-world coordinate system.
Relative accuracy is achieved through correct technique and loop closure. It does not require RTK, GCPs, or any external reference. It is the baseline result from any properly executed scan with genuine loop closures throughout the route. In environments with poor scene structure, long linear paths without loops, or technique errors, relative accuracy degrades regardless of which device is used.
Absolute Accuracy
Absolute accuracy is how closely the point cloud registers to a real-world coordinate system. It requires external control: RTK, PPK, or surveyed ground control points. Technique alone cannot produce absolute accuracy, no matter how well the scan is executed.
The 3 cm RMSE absolute accuracy figure (L2 Pro, K2, K1) applies when RTK or GCPs are used correctly. On the K2 the published figure carries a stated maximum of 5 cm, so quote the range rather than the headline number when a client is holding you to a tolerance. None of these figures describe a pure SLAM scan. A scan without any georeferencing has no defined absolute accuracy because it has no connection to a coordinate system.
The Two Sources of Elevation Error
Elevation error in SLAM scans comes from two independent sources that must be addressed separately.
IMU Leveling Error
The SLAM system uses the IMU to determine level. Over large areas, a small angular error compounds into meaningful elevation discrepancy. The K2 leveling figure is 0.05 degrees. The L2 Pro publishes tighter angular figures: horizontality 0.015 degrees with RTK, verticality 0.03 degrees.
RTK eliminates this error by providing an absolute vertical reference. Surveyed GCPs with Z coordinates also eliminate it.
SLAM Accumulated Drift
SLAM systems accumulate positional error in X, Y, and Z over distance due to the absence of absolute constraints. Loop closure distributes and reduces this error. Ground control points constrain it. The denser the control point network, the smaller the residual drift error.
RTK alone does not fully address SLAM drift across long trajectories.
Why the hybrid approach (RTK plus GCPs) produces the best results for demanding projects. RTK provides continuous coordinate alignment and eliminates IMU leveling error. Surveyed GCPs also eliminate IMU leveling error and constrain accumulated SLAM drift at discrete anchor points across the dataset. Together they provide redundant leveling correction and the tightest overall constraint on both error sources.
Loop Closure, and Stopping the Scan Correctly
Loop closure is the mechanism by which SLAM corrects accumulated drift. When the scanner returns to a previously visited area and the algorithm detects the match, it measures the accumulated error and distributes the correction back across the trajectory. Loop closure requires two conditions:
- You physically return to a position you have already scanned.
- Your viewing angle when you return is within approximately 40 degrees of your original angle at that location.
Loop closure improves relative accuracy. It does not provide absolute accuracy. A perfectly looped scan with no georeferencing still has no connection to a real-world coordinate system.
The Final Loop Closure Is a Procedure, Not a Location
The last loop closure of a session is the one that corrects everything the session accumulated, and it is also the one most often thrown away. Walking back to the start point and pressing stop does not complete it. The device needs a few seconds of stationary observation at the recognised position to run the match and redistribute the correction.
Skipping the stand-still is a documented cause of point cloud layer separation. Layering appears as a wall reconstructed as two offset surfaces, or a floor and ceiling that do not line up between the start and end of the route. It is not recoverable by reprocessing in most cases and it fails a job that was otherwise scanned correctly. Ten seconds of standing still is the cheapest accuracy insurance in the workflow.
Realistic Accuracy Outcomes by Method
The table below shows the accuracy outcomes achievable with each control method, the conditions required to achieve them, and the factors that degrade each result. These ranges reflect field reality, not idealized test conditions.
Measure Point is the answer to the RTK-denied transition. Underpasses, loading docks, tunnels, and the first rooms inside a building are where RTK drops and where an absolute coordinate is still needed. On the L2 Pro, walking a qualifying L-shaped route in Fixed status before entering the denied area carries absolute positioning into it, at 5 cm within the first 164 ft (50 m). The route must be walked before you lose signal. There is no way to add it afterwards.
Method Selection by AEC Project Type
The right accuracy method depends on the deliverable requirement, not the device capability. Choose based on what the client or downstream workflow actually needs, and confirm that requirement before leaving for the field. The control strategy cannot be changed after the scan.
As-Built Documentation for BIM
BIM deliverables are used for design coordination and construction. Dimensional accuracy drives fit and clash detection. Absolute coordinate registration enables multi-session dataset alignment and integration with survey control.
Minimum requirement: GCPs with surveyed coordinates, at least three per scan, evenly distributed and not in a straight line. Maximum interval is 328 ft (100 m) on the L2 Pro and 164 ft (50 m) on the K2. Verify GCP residuals in LixelStudio before delivery.
When to use hybrid: Large facilities where the scan spans multiple floors or sections. RTK constrains the elevation across floors; GCPs constrain drift within each floor.
Pure SLAM is insufficient if the BIM model will be used for construction coordination or clash detection. Dimensional errors compound across large datasets.
Legal Survey or Cadastral Work
XGRIDS mobile SLAM is not a replacement for licensed survey instruments for legal boundary work. The system can provide supporting documentation and area measurements, but legal survey requirements vary by jurisdiction.
If XGRIDS data is used as supporting evidence: Use hybrid RTK plus GCPs, run an accuracy check against independently surveyed checkpoints, and retain the accuracy report as documentation. See Section 7.
Accuracy context: The 3 cm RMSE absolute specification can support many legal documentation requirements, but confirm with the licensed surveyor of record before committing the data for legal purposes.
Do not misrepresent mobile SLAM data as survey-grade without independent verification.
Construction Progress Monitoring
Progress monitoring requires repeatable registration between scan sessions more than it requires the highest possible absolute accuracy. What matters is that each session aligns to the same coordinate reference so that differences represent actual construction change, not scan-to-scan registration error.
Key principle: Use the same control strategy and the same physical GCP locations across every session. Changing the control network between sessions breaks the registration consistency.
Permanent GCP monuments are worth installing on long-term projects. They eliminate re-survey costs and guarantee consistent registration across dozens of scan sessions.
RTK alone is acceptable for open-air construction sites where Fixed status is reliable and consistent.
Visualization and Client Delivery (3DGS)
3D Gaussian Splat deliverables via LCC Studio are primarily visual. Clients viewing the model in a browser are not measuring dimensions against a coordinate system. Absolute georeferencing adds cost and complexity that typically does not improve the deliverable for visualization purposes.
When absolute coordinates matter for 3DGS: 3D Tiles export requires absolute coordinates, so a WebGIS or Cesium deliverable forces the decision before processing. Multi-session work also benefits from a shared coordinate reference.
PortalCam-only projects: The PortalCam does not produce a point cloud and does not output absolute coordinates. It does use control points, which connect segments during Map Fusion. Its internal geometry is adequate for visualization, space planning, and client presentation, and is not a substitute for measured deliverables.
Industrial Facility or Large Indoor Site
Large indoor facilities are the most demanding accuracy scenario for mobile SLAM. RTK signal is unreliable indoors. Long trajectories accumulate drift. Multi-floor coverage introduces vertical error from IMU leveling that compounds across levels.
Control point spacing for large indoor facilities: Published maximums are upper limits, not targets. For facilities where high dimensional accuracy matters, space control points at 164 ft (50 m) or less with the L2 Pro, and at 80 to 100 ft (25 to 30 m) with the K2 or the legacy K1.
Multi-floor projects: Place GCPs on each floor and include points visible from stairwells or open voids that span floors. This constrains vertical drift across the building height.
Walk-and-stop technique at control points improves accuracy. Set the scanner down gently at the marker, wait for the app to confirm the point was added, then circle the point once or twice before continuing. After marking the last point, wait at least 15 seconds before ending the recording.
Outdoor Site with RTK Coverage
Open outdoor sites with unobstructed sky view are where RTK performs most reliably. If the environment supports a sustained Fixed satellite lock throughout the scan, RTK alone can achieve the published 3 cm absolute accuracy specification.
Verify Fixed status before beginning: Confirm more than 10 satellites and a Fixed status (not Float, Single, or None) in LixelGO. Walk at least 33 ft (10 m) while Fixed before starting the scan route. Coordinate transformation in processing needs a continuous 33 ft (10 m) of Fixed data to become available at all; without it the GNSS page reports the transformation as unavailable and there is nothing to recover.
Hold enough valid data: The L2 Pro needs more than 100 valid RTK data points for coordinate conversion. The K2 minimum is more than 20 valid points with a scan area over 33 ft (10 m), and that minimum produces reduced accuracy rather than the published figure.
Keep unfixed runs short: On the L2 Pro, keep any continuous unfixed section under 328 ft (100 m). On the K2 and K1, keep it under 164 ft (50 m). These are different numbers for the same reason the control point intervals differ.
Antenna and device tilt: On the L2 Pro, keep the device within 20 degrees of vertical while walking and the RTK antenna within 10 degrees. On the K2 the RTK antenna is built in, so RTK mode imposes the tighter limit on the device itself: keep it within 10 degrees of vertical, which is stricter than the general posture rule. Exceeding these invalidates the fix record even when the LED stays green.
When to add GCPs to outdoor RTK scans: If the site includes areas under tree canopy, near tall structures, or in urban canyons where RTK signal may drop or degrade, add GCPs in those zones. Do not assume RTK is maintaining Fixed throughout the entire scan.
The accuracy method must be chosen before the scan begins. RTK must be configured and confirmed as Fixed before scanning starts. GCPs must be physically placed and marked in LixelGO during the scan. There is no way to apply georeferencing to a scan after the session without control that was captured during it. A scan collected without any georeferencing cannot be retrospectively tied to a real-world coordinate system.
Seven-parameter transformations: the scale value must be in ppm. If a surveyor supplies a Bursa-Wolf scale factor expressed as k, convert it before entering it: ppm equals (k minus 1) multiplied by 1,000,000. Entering k directly produces a point cloud that looks blurred or subtly distorted rather than obviously wrong, so it is usually discovered after delivery. Check the unit every time, even when the transformation has worked on a previous job for the same client.
Verifying and Documenting Accuracy
Everything above describes accuracy you should be able to achieve. None of it proves what you achieved on a given job. Accuracy Check in LixelStudio compares checkpoint coordinates in the processed cloud against independently surveyed true values and produces a report you can hand to a client. On any job where a tolerance is contractual, run it before delivery.
Plan Verification Before You Leave the Office
Checkpoints are separate from control points. Control points constrain the solution; checkpoints test it. Using the same points for both tells you only that the software applied your control, not that the result is accurate. Survey a set of checkpoints that were not used as control, and place targets on them.
What the Field Work Requires
- Use the supplied circular reflective target paper, 4 cm. A different shape or size will not be recognised automatically and the check cannot run. There is no substitution.
- Scan around each target, or pause at it, so the target surface accumulates enough points for automatic extraction.
- Capture at least three checkpoints. Fewer will not produce a result.
- Walk a verification route of at least 328 ft (100 m) that is not a straight line. A short or linear route does not exercise the trajectory enough to test it.
Accuracy Check Parameters
What the Report Gives You
The check calculates the coordinate difference, planar error, and elevation error for every checkpoint pair, reports the maximum, minimum, and average of each, and gives the mean square error for plane and for elevation. The exported report is written to a report folder inside the project directory.
What to hand the client. The report is the difference between saying the scanner is specified at 3 cm and showing that this scan met 3 cm on this site. On progress-monitoring work, run the check on every session and keep the reports together; the trend across sessions is what tells you whether a change in the model is real construction or drift. On a job where the tolerance is contractual, the report is the document that closes the question.
Manual checkpoint selection changes what you are measuring. If the targets were not placed during acquisition, checkpoints can be picked by hand in the cloud, but you are then testing your ability to identify a feature centre rather than the scanner's accuracy at a surveyed point. Use it for a rough sanity check, not as evidence for a client.
©2026 Alpine Reality Capture LLC • XGRIDS Pro Guide™ • Site Disclaimer