3.5 Data Centers

Section 1

What Makes Data Centers Hard to Scan

Geometric repetition. Every aisle between rows looks nearly identical. The LiDAR sees the same wall-to-wall geometry repeating at regular intervals for the full length of the hall. Without additional reference features, the system has limited material to distinguish one position from another, and drift accumulates faster than in environments with varied geometry.

Low visual texture. Rack faces, raised floors, and metal ceilings provide limited visual feature variation. The visual SLAM component relies on texture to match frames. In the worst case, a freshly painted server hall with identical black rack fronts on both sides provides almost no visual anchoring.

No GPS. RTK cannot initiate a Fixed solution inside a building. A Fixed solution established outdoors before entry carries in and provides absolute coordinate coverage for a limited distance of interior scan path from the threshold: approximately 330 ft (100 m) for the L2 Pro, or 165 ft (50 m) for the K1. This is viable for facilities with accessible loading docks or equipment entries. Loading docks with partial overhead cover and adjacent building walls are marginal GNSS environments. The Survey-Grade RTK module's higher gain antenna (5.5 dBi vs 2.8 dBi on the Standard module) improves the likelihood of achieving and holding Fixed status at these entry points. Confirm Fixed status in LixelGO regardless of which module is installed before crossing the threshold. Beyond these distances of interior scan path, or where no outdoor sky view is available near an entry, absolute georeferencing requires ground control points with surveyed coordinates obtained before the scan begins using a total station referenced to an exterior control network, or by occupying each marker position with a GNSS receiver where sky visibility permits. See Module 4 for the full RTK and GCP strategy.

Live equipment. Scanning around powered racks limits where targets can be placed. The environment may include small vibration sources and heat gradients near hot-aisle vents that introduce minor scan noise.

Section 2

Scanning Technique

Modular Capture by Pod or Section

Break capture into logical sections rather than scanning the entire floor in a single continuous session: 2 to 4 rows per session, or by structural pod boundaries if the facility uses that layout. Shorter sessions reduce per-session drift accumulation, allow targeted rescans of only the zones that changed, and keep individual file sizes within reliable processing limits.

Scan the main cross-aisle or backbone corridor first, then branch into individual aisles from it. Every aisle entry and exit back to the corridor creates a loop closure, continuously correcting drift rather than letting it accumulate across the full session length.

Aisle Pattern

Walk each aisle in a serpentine or ladder pattern rather than a straight centerline pass. A serpentine path weaves from hot-aisle to cold-aisle side at a steady pace, capturing rack faces from opposing angles on adjacent passes. This produces overlapping coverage that supports more reliable tracking than a single midline pass through a uniform-geometry corridor.

At each end of every aisle, walk a small figure-eight or loop before reversing. These intersections are the primary loop closure opportunities in the data center environment. Use every one of them rather than turning sharply and reversing course.

Height Variation

Vary the device height between passes where practical. A pass at chest height followed by a pass at above-head height captures top-of-rack geometry and overhead cable tray routes that a single-height pass misses. Transition between heights gradually over several feet of travel rather than in a single step. Abrupt angle changes give SLAM less stable data than smooth, continuous transitions.

Section 3

Target Strategy in Data Centers

The L2 Pro and K1 ship with reflective sticker targets (referred to as GCP Collection Plates in the packing list). The standard marking procedure is to place the target on a stable surface, set the scanner down on it during the scan, and mark the control point in LixelGO. They work in any area where the surface is accessible and the scanner can be briefly set down.

Data centers present constraints that sticker targets alone may not fully address. Raised floors, dense cable management, live equipment in aisles, and the need for targets at varying heights to create 3-dimensional control networks all push beyond what a single-height-plane control network can provide.

XGRIDS Sticker Targets: Standard Procedure and Practical Limits

The reflective sticker targets included with the scanner are used as GCP markers. The documented procedure is: place the target on a flat, stable surface; during the scan, circle the sticker 1 to 2 times before setting the scanner down on it; mark the point in LixelGO; then lift the scanner and continue. The practical limit in a data center is that this procedure requires floor access and a stable surface where the scanner can be set down, which restricts placement options in live rack aisles and keeps the control network at a single height plane unless the floor layout provides sufficient geometric variation.

Professional Targets That Extend the XGRIDS System

3 target types are standard in professional scanning programs and work alongside XGRIDS workflows. All 3 can be used as GCP markers whose surveyed coordinates are loaded into LixelStudio during processing, or as SLAM anchors whose geometry the system recognizes during scan registration.

Building a Permanent Control Network

For facilities with recurring scan programs, a small permanent target network is the highest-value investment. Fix sphere mounts or adhesive bases at a limited number of positions in main access corridors, pod intersections, stairwells, and representative mechanical rooms. Survey those positions once with a total station to establish their absolute coordinates. Every future scan campaign places targets on the same mounts, marks the same point names in LixelGO, and loads the same coordinate CSV in LixelStudio. Registration becomes consistent and repeatable across years without resurveying.

The removable magnetic targets on rack ends and containment framing serve a different purpose: they are SLAM anchors and session-to-session alignment aids, not absolute control points. Use the permanent floor network for absolute coordinate registration and the removable magnetic targets for local loop closure and change detection alignment.

Section 4

RTK Strategy for Data Centers

RTK cannot initiate a Fixed solution inside a data center building. A Fixed solution established outdoors before entering carries into the facility and provides absolute coordinates for a limited distance from the entry point: approximately 330 ft (100 m) for the L2 Pro, or 165 ft (50 m) for the K1. Accuracy degrades with distance from the Fixed loss point: the L2 Pro maintains approximately 2" (5 cm) absolute accuracy within 165 ft (50 m) and approximately 4" (10 cm) within 330 ft (100 m). For most single-pod or single-floor scans entered from a loading dock or main entry with outdoor sky view, this is a viable georeferencing approach that requires no separate survey work for the threshold control points.

NTRIP Is Required

The L2 Pro RTK module receives correction data from a CORS network via NTRIP over a mobile data connection. Without an active NTRIP connection, the module cannot achieve Fixed status outdoors and there is no coordinate frame to carry indoors. Confirm mobile data coverage at the facility entry before relying on RTK for the session. If coverage is unavailable or unreliable at the site, PPK with a base station or surveyed GCPs are the alternative paths.

Single-Entry Workflow

For a facility where the scan area is within the device's maximum unfixed distance of a single outdoor-accessible entry: achieve Fixed status and complete the L-shaped initialization route outdoors (see Module 4). Place GCP sticker targets at the entry threshold, 1 just outside and 1 just inside. Walk continuously from outside to inside. Mark the threshold GCPs in LixelGO while the RTK carry-in is still active. Continue the interior scan from that anchored starting point.

Multi-Entry Workflow for Large Facilities

For large data center halls where the interior extends beyond the device's maximum unfixed distance from any single entry, plan multiple scan segments each initialized from a different building access point with outdoor sky visibility. Loading docks, emergency exits, and equipment bay doors distributed around the facility perimeter each become an RTK initialization point. Each segment carries its own absolute coordinate frame from its respective entry. In LixelStudio, Map Fusion registers those segments together using shared GCP targets placed in the overlap zones between them.

Section 5

Rescan Alignment

The value of periodic scanning compounds when each rescan aligns precisely to the baseline. A consistent target network and consistent scan routes make this repeatable without manual registration effort each time.

• Re-occupy the same control points across sessions. Return targets to the same physical positions used in the baseline scan and mark them with the same point names in LixelGO. This applies to floor sticker targets, magnetic checkerboard plates on rack ends, and sphere mounts in corridors equally
• Start and end loops in unchanged areas. Begin and end each rescan session in a corridor or zone that was not modified since the baseline. The geometry in that area matches the original scan and provides initial and final loop closure reference that anchors the new session to the existing dataset
• Scan only the sections that changed where possible. A targeted rescan of 2 modified rows, anchored to unchanged corridor geometry on both ends, produces a precisely aligned result without rescanning the full hall. Map Fusion connects the new segment to the existing baseline using shared overlap area and control point names
• Overlap between sessions must be at least 50 ft (15 m) of shared scanned area. Plan for 65 to 100 ft (20 to 30 m) to account for field adjustments that reduce actual overlap below the minimum threshold

Section 6

Scan Cadence

Section 7

Advanced Trajectory Design and QA/QC

Standard serpentine technique keeps SLAM stable within a single aisle. At pod scale and above, trajectory design becomes the primary accuracy control. The decisions made before the scan starts determine whether the registered dataset is usable or requires recapture.

Snake Patterns Across Pods

Long straight hot or cold aisles present nearly identical geometry to the SLAM engine for their entire length. Drift accumulates in proportion to the feature poverty of the path. The fix is lateral variation built into the route.

For each pod, use a snake pattern rather than a sequential aisle-by-aisle approach. Enter Cold Aisle 1 from the main corridor. Walk to the far end. Cross through the adjacent hot aisle or a cross-aisle into Cold Aisle 2. Walk back to the corridor entry. Repeat for Cold Aisles 3, 4, and so on. Each cross-connection introduces geometry the system has not seen from that angle, and each return to the corridor creates a loop closure that corrects the accumulated drift from the previous run. A pod captured as 6 connected aisle passes with 5 lateral cross-connections is dramatically more stable than the same pod captured as 6 sequential straight passes.

Corridor Anchors at Pod Boundaries

Design every pod trajectory to start and end at the same corridor segment, and to return to that corridor at regular intervals across the session. Main corridors, pod entry doors, structural columns, fire panels, and wall-mounted signage give the SLAM algorithm substantially more to work with than rack aisles do. These features serve as natural anchor points that bound drift within the pod and prevent it from propagating into adjacent pods during Map Fusion.

Do not design a trajectory that exits 1 pod and enters the next without returning to a feature-rich corridor segment first. The corridor intersection between pods is a mandatory waypoint, not an optional one.

Aliasing in Multi-Pod and Hyperscale Halls

In facilities with many identical pods, the primary registration risk is aliasing: the SLAM engine or Map Fusion treating 1 pod as another because their geometry is indistinguishable. This does not produce a subtle error. It produces a structurally wrong dataset where entire rows are overlaid on incorrect positions.

Capture each pod as its own SLAM loop anchored to the main corridor. Do not rely on 1 continuous trajectory that walks through multiple identical pods. At each pod entry, ensure the route includes at least 1 distinctive feature cluster: a specific door arrangement, fire device, column position, or signage pattern that differentiates this pod from adjacent ones. These become the registration fingerprints that prevent pods from being merged incorrectly during processing. Use short, intentional overlapping passes in shared corridors between pods to enable registration, and keep the bulk of each pod's geometry constrained to its own anchors and control points.

Hybrid SLAM with Static Scan Anchors

For large facilities where corridor-to-corridor absolute accuracy matters for BIM coordination or campus-level registration, a hybrid approach produces results that SLAM alone cannot reliably achieve. Place a small number of static laser scans in main access corridors, pod intersections, stairwells, and mechanical rooms. These positions have rich geometry, are feature-stable across rescan visits, and are exactly where SLAM drift tends to compound over long sessions.

Run SLAM loops that start and end on these static scan positions. During registration, constrain the SLAM trajectory to the static anchors so that pod-level drift is clamped to the corridor geometry rather than accumulating freely. The result is a dataset where white-space coverage comes from SLAM efficiency and absolute accuracy comes from the static anchor network.

Overhead and Under-Floor Capture

Cable trays, busways, and overhead power distribution are often the primary deliverable for data center renovation planning, and they require deliberate route design to capture correctly.

For overhead infrastructure, walk both sides of major cable trunks along main corridors to give the system multiple viewpoints on the same overhead geometry. Insert short transverse segments at regular intervals, crossing from 1 side of the corridor to the other, to avoid long collinear paths under identical tray runs. A single midline pass under a uniform overhead tray provides almost no parallax and captures only the underside of the geometry. Passes from both sides at staggered intervals capture tray depth, branching, and the geometry above.

Where raised-floor tiles can be safely removed for under-floor access, run short SLAM loops from 1 opening to another and back. Tie these loops to the overhead session during registration so that under-floor piping and power routing aligns accurately with the white-space model above.

Containment and Glass Environments

Cold aisle containment with glass or clear acrylic panels affects both the LiDAR and the visual cameras. LiDAR pulses produce unreliable returns off glass: a mix of reflections from the panel surface, false returns from objects behind it, and scattered noise. The visual cameras see reflections at some angles and see through the glass at others, making feature matching unreliable near the panels. In a narrow containment corridor (typically 4 ft / 1.2 m wide) with glass on both sides, these effects are concentrated and harder for SLAM to compensate for.

Scan both sides of containment. From inside, the scanner has direct line of sight to rack fronts, patch panels, cable management, top-of-rack geometry, and immediate overhead structure without glass between it and those surfaces. From outside, the scanner captures the containment framing, door hardware, the glass panels as geometry, and the surrounding hot aisle corridor including overhead cable trays, busways, and equipment on the rear of racks. The 2 perspectives fill each other's gaps.

Containment door transitions. Containment doors are typically spring-loaded or magnetically held and close automatically. The standard half-open door technique from Section 3.3 does not apply because the door will not stay in position. Instead, approach at a steady walking pace and pass through without stopping in the doorway. If you need the door in its closed state for the dataset, scan it from outside the containment before entering, then pass through for the interior scan.

Movement in glass-lined aisles. Walk at a steady, continuous pace. Do not stop or make sudden direction changes inside a glass-lined containment corridor. Stopping leaves the scanner stationary in a space where every nearby return is contaminated by reflection. Sudden turns shift the reflection pattern on all surrounding glass surfaces at once. Both can cause trajectory errors that affect everything scanned afterward and are difficult to recover from without recapture. Treat containment aisles as a single continuous pass: enter, walk the length at 1.5 ft/s (0.5 m/s), and exit.

On-Site Visual Checks Before Leaving

The checks that can be performed on site are visual inspections of scan coverage and obvious drift indicators visible in the LixelGO preview during or immediately after scanning.

A problem caught on site costs one rescan session. The same problem discovered after the team demobilizes costs a return visit. Dimensional accuracy verification, control point alignment analysis, and full registration quality assessment are performed during post-processing in LixelStudio, not in the field.