3.5 Data Centers

Why Data Centers Are Different

The geometry is dominated by rack rows, each visually identical to the next. Floor markings and structural columns are the only natural features that distinguish one position from another along an aisle, so SLAM cannot rely on rack geometry to know where it is.

These sites are also scanned repeatedly, often quarterly, to track equipment changes and as-built drift. Every visit must align with the last and with every future one. That places long-term consistency requirements on target placement, route, and reference geometry that one-off scans do not have, so the technique here is more rigid. Departures compound across visits and degrade the whole engagement.

Route Strategy Through Aisles

Speed and Posture

The speed limits are not the same on all three devices, and a data hall sits at the slow end of every band because rack faces on both sides create a low-feature environment throughout.

• Hold posture consistent along the row. Rotating mid-aisle to face one rack changes the view of both rows at once and introduces alignment artifacts.
• Keep the LiDAR field of view clear. No object may block more than 50 percent of the field of view within 3.3 ft (1 m) for sustained periods. In a narrow aisle that includes your own body and anything you are carrying, so hold extra gear behind the scanner relative to your direction of travel.
• Stay at least 1.6 ft (0.5 m) off the rack faces, and 1.6 to 3.3 ft (0.5 to 1 m) on the K2. Cold aisles narrow enough to force you closer than that are a coverage problem, not a speed problem.
• Lens orientation follows aisle width. In open areas of the hall, one lens faces left and one right of your direction of travel. In aisles narrow enough to count as a corridor, rotate the device 90 degrees so one lens faces the direction of travel.
• Aim for 60 to 80 percent overlap between frames. That is what the slow pace buys you, and it is the number the reconstruction actually depends on.

Aisle Coverage

• Walk every aisle. Do not skip cold aisles if hot aisles are done. Skipping produces weaker 3DGS and incomplete point cloud on the unwalked side of every skipped row.
• Vary scanner height between aisles. One aisle at chest height, the next at overhead height, changing gradually over several feet. This gives each rack row at least two viewing heights. Alpine practice, not a published figure.
• Loop back through the cross-aisle at the end of every row to reset accumulated drift before the next corridor.
• Close the hall loop and stand still before you stop. Return to the point you initialized at, hold the device steady facing the original direction, and wait before ending the recording. Ten seconds on the K2, several seconds on the L2 Pro and PortalCam. Skipping it discards the loop closure that corrects the whole session. Full procedure on 3.2 Route Planning.

Three-Pass Coverage for High-Value Halls

For detailed deliverables such as capacity planning, equipment audits, and change tracking, three passes in a single continuous session produce substantially better coverage than one thorough walk. Each pass has a different job, and the third is the one people skip.

• Pass 1, perimeter loop. Walk the walls of the hall slowly with the lens pointed across the room toward the far side. Stay 1.6 to 3.3 ft (0.5 to 1 m) off the wall at 1.0 to 1.6 ft/s (0.3 to 0.5 m/s). This records the overall structure of the space and gives stable side-view overlap between frames. Pause longer at doorways.
• Pass 2, grid traversal. Move into the body of the hall and cross it two to three times in each direction to form a grid, lens at eye level facing your direction of travel. This captures the depth relationships between rows and is what rescues weak-texture surfaces, which in a data hall means blank rack panels, containment walls, and open floor.
• Pass 3, ceiling and floor. Walk the same routes again with the lens angled up, then angled down, within the tilt limit for your device. Up recovers overhead cable tray, busway, lighting and containment roof. Down recovers raised-floor tiles, perforation patterns, under-rack cabling and floor markings. This is the pass that makes a capacity model usable.

All three passes are one continuous session. Stopping between them creates artifacts at the joins and, because no XGRIDS scanner can resume from a breakpoint, turns one session into several that must be fused. Keep carry height consistent within a pass and change it gradually between passes.

Anchor Target Placement

An anchor target is a high-contrast marker that gives SLAM something reliable to match in an environment where the geometry repeats. It is not a survey point. Attach a surveyed coordinate to it in processing and the same physical target becomes a ground control point instead. In a data hall you almost always want both roles filled, and often by the same markers.

• Space targets every 100 to 165 ft (30 to 50 m) along the scan path, about one per 3 to 5 aisles. Alpine practice, adjusted to how repetitive the hall actually is.
• Any flat, high-contrast, non-reflective marker works. The K2 ships with 30 adhesive reflective stickers suited to this, and they adhere to any flat surface without needing a ferromagnetic mounting point.
• Confirm adhesive is permitted before the visit, not on arrival. Adhesive is the documented mounting method, so an adhesive prohibition is a scope problem rather than an inconvenience. Many operators restrict adhesives on rack panels and some treat residue as a chargeable item. Agree target type, placement zones, and removal with the facility in advance.
• Place each target where the scanner can approach within 3 to 6 ft (1 to 2 m) and circle it during the scan. Targets behind equipment lose most of their value.
• Keep targets off hot-aisle vents and active airflow paths. Turbulent air near cooling exhaust vibrates a marker and can shift it mid-scan. Choose flat panel faces away from airflow. Published scene-preparation guidance is to secure lightweight items near air-conditioning vents before capture, and a loose target is exactly that.
• Mark every target in the field app during the scan. LixelGO on the L2 Pro and K2, LCC Scan on the PortalCam. An unmarked target is visible in the point cloud but is not used for SLAM correction or georeferencing in processing.
• Place targets in the same positions as the original scan. For recurring rescans, record locations permanently with floor coordinates or photographs.

Georeferencing Without GPS

Most white space has no GPS signal, so georeferencing depends on carried-in RTK and ground control points. The PortalCam produces no georeferenced output at all, so a data hall job with an absolute-coordinate deliverable needs an L2 Pro or a K2.

Indoor RTK Reality

• Initialize RTK outdoors and carry it in. Reach Fixed at the entrance under clear sky, walk at least 33 ft (10 m) while Fixed, then carry the scanner into the white space.
• RTK drops at the first overhead structure. Indoors the scanner falls out of Fixed almost immediately, and the indoor solution rides on the last Fixed position.
• Unfixed-run tolerance differs by device. Keep any continuous stretch without a fix under 328 ft (100 m) on the L2 Pro and under 164 ft (50 m) on the K2. Beyond those, expect coordinate drift.
• Past the device's tolerance, control points are required. RTK alone is insufficient for whole-facility georeferencing beyond compact server-room scale.

Ground Control Point Strategy

• A control point needs surveyed coordinates to georeference anything. Without them it is an anchor target with a known location: useful to SLAM, useless for absolute coordinates.
• Place control points every 164 ft (50 m) for the K2 and every 328 ft (100 m) for the L2 Pro. Closer improves accuracy and adds redundancy if one fails to mark cleanly.
• Never place them in a straight line. Collinear control does not constrain rotation about that line, so the model can twist around it while every residual still looks acceptable. Distribute them as a polygon across the hall.
• A multi-session hall needs at least 3 absolute control points across all sessions, not collinear, for LixelStudio to perform the coordinate transformation. This is a project-level requirement, not a per-session one, and it is easy to miss when sessions are planned independently.
• Use the same coordinate system as the carried-in RTK base. Mixing systems forces a transformation in processing and loses precision.
• Document positions and the coordinate system in the deliverables. Datum, projection, zone. Future rescans must reuse them.

Permission and Access

• Confirm control point placement permission before the scan. Many operators prohibit markers on raised-floor tiles, rack surfaces, or structure. Identify acceptable zones in a pre-scan walkthrough.
• Adhesive removal is mandatory under many agreements. Remove sticker targets and tape at session end and photograph the removed state.
• Photography and floor-plan access often require escort and time limits. Plan documentation inside the access window.

Recurring Rescans

The original scan sets the baseline; each rescan captures change. For change detection to work, every scan must align with the baseline to sub-inch precision. Match the baseline on three things.

Targets

• Same positions as the baseline. Record them with floor coordinates, rack identifiers, or annotated photographs so any operator can reproduce them.
• If a position is gone because a rack moved or was replaced, place a substitute at the closest equivalent, document the change, and carry the note forward to the next visit.
• Photograph each placement with rack and target ID before the scan. Photos guide future placement better than written coordinates do.

Route and Boundaries

• Walk the same aisles in the same order, at the same speed and posture. A different pace changes point density, which shows up as apparent change where nothing changed.
• Start and end at the same locations as the baseline, within a few feet, and follow the same stop procedure.
• Match the baseline's session boundaries: same count, same locations, same overlap zones.
• Every session must share at least 50 ft (15 m) of geometry with the session it merges with, and 50 to 100 ft (15 to 30 m) is the recommended band. Place the overlap in a feature-rich area such as a cross-aisle intersection or a plant room, never a blank containment run.
• Keep each capture under 20 minutes where the point cloud will be joined in LixelStudio. A longer capture cannot be fused at all, and on a large hall this limit, not the battery, is what decides where the boundaries fall. Plan them before the visit.
• Keep the same reconstruction settings between visits, including the layered optimization state from Section 1. Changing a setting between generations changes the output in ways that read as site change.

Before You Leave Site

Access windows are restricted and return visits are expensive. This checklist catches the common failures before the window closes.

• Scan stopped cleanly: the closing loop was held before stopping, and the indicator reached solid green before power-off. No mid-session error.
• Duration matches plan: within 15 percent of the estimate for the area covered. A large gap is worth investigating before leaving.
• Preview shows no aisle kinks or pod-boundary discontinuities. Check at the cross-aisles between every pod section. Gradual separation between passes over the same aisle is drift and is usually recoverable. An abrupt jump is not, and it means a recapture now rather than a revisit later.
• Every aisle is present in the preview, shown as parallel rack rows.
• All three passes completed where the deliverable calls for them, including the ceiling and floor pass.
• Targets marked in the field app during the scan, every one that was placed.
• Targets and tape removed per the operator agreement, with the removed state photographed.
• Control point positions documented with the survey method used and the coordinate system named.
• Photo log captured: each placement with rack and target ID, plus notable equipment changes since the baseline.