SHARE C10-32 Pro

Drift comes from feature-poor geometry: long straight corridors, blank tunnel walls, open flat ground and repetitive bays where consecutive frames look identical. Walk slower through those stretches, close a loop back to a feature-rich area every few hundred metres, and break a very long traverse into segments you register afterwards. Loop closure is what pulls accumulated error back, so plan the loop before you start walking, not after.

Point density is a function of how long the sensor dwells on a surface, so walking faster thins the cloud and increases the chance of a registration gap. A steady walking pace suits open ground. Slow to a shuffle through doorways, stair transitions, tight plant and any turn sharper than about 90 degrees, because fast rotation is the most common cause of a lost lock.

They appear as smeared points along their path and can corrupt the SLAM solution if they fill the field of view. Keep moving objects out of the near field, and avoid walking directly behind a vehicle. Most moving-object noise filters out in processing, but a forklift that occupies half the frame through a doorway can break the lock, not just add noise.

It changes nothing in the data. The sensor solution is identical either way, so you can move between the two mid-walk without stopping the scan. Handheld is right for most work and better for reaching over an obstacle or scanning under a structure. The chest mount takes the weight off your arms on a full day of capture and holds the sensor at a consistent height, which makes ground classification easier later.

Relative accuracy is better than 1 cm and point cloud thickness stays within 5 mm on a clean walk with good loop closure. Absolute accuracy is better than 5 cm with an RTK fix, from H 0.8 cm + 1 ppm and V 1.5 cm + 1 ppm positioning. Those figures assume sky view for the RTK and a well-planned path. A blind traverse with no loops will not hold them.

The SLAM solution keeps running, so you still get correct relative geometry, but the section captured without a fix has no absolute reference and can shift when you georeference. Either re-acquire a fix and close a loop back through the surveyed section, or tie that section to known control points in processing. For work going underground or under cover from the start, plan control targets before you scan rather than hoping for a fix.

Three options, and you can combine them. Built-in RTK against a network like AUSCORS georeferences the walk live. PPK post-processes the same trajectory against a base log if you would rather not rely on a live link. Surveyed control targets placed through the site let you transform and check the cloud against a known network afterwards, which is what most acceptance criteria will ask you to demonstrate.

Check three things on the tablet before you pack up. Coverage, by looking for holes rather than assuming the walk covered them. Loop closure, by confirming the trajectory returned to its start without a visible step. And a spot dimension against something you can tape measure, like a door opening or a column spacing. Uncaptured area cannot be recovered in processing, so a second mobilisation is the only fix.

Glass, mirrors, still water and polished metal either return nothing or return a false reflection behind the surface. Thin cables, mesh and fine railings return unstable points. Capture the primary structure with the scanner and measure the problem surfaces manually, or apply temporary markers or film where the surface has to appear in the cloud. Approach reflective facades at an angle rather than square on.

Yes, significantly. The published range figures are quoted at 80 percent reflectivity, which is a light, dry, diffuse surface. Fresh asphalt, wet ground, dark rock and black plant sit closer to 10 percent, where usable range drops to around 80 m. Plan pit and stockpile walks against the low-reflectivity figure, not the headline one, or you will find gaps at the far edge in processing.

A full day of capture fills a meaningful share of the 1 TB drive, and processing is memory-bound rather than CPU-bound. Budget 64 GB of RAM as a working minimum for large sites, fast local NVMe rather than network storage, and a CUDA-capable NVIDIA card if you intend to produce 3DGS output. Processing large clouds over a network share is the most common self-inflicted bottleneck.

You get measured spatial data, exported as LAS, LAZ, E57 or PLY. That is a survey product, not a model. It imports into CAD, Revit and SketchUp as a reference underlay you view, measure and model against. It does not become an editable BIM model, and it is not a mesh until you run reconstruction. Automated extraction produces drafts that a person still has to finish.

Yes, and on large sites you should plan for it rather than attempting one heroic traverse. Give consecutive walks a generous overlap through a feature-rich area, ideally 20 to 30 metres of shared geometry with real structure in it, and start and end each walk at the same identifiable point. Georeferenced walks register on their coordinates. Walks captured without a fix register on shared geometry alone.

Yes. The dual-battery system is designed for hot-swap, so you replace one pack while the other carries the load and the scan continues. Stop walking and stand still while you swap, because the motion solution is more fragile during the changeover. Two packs give about 180 minutes, so a full day means planning charge cycles rather than carrying one spare.

Airborne dust and rain both return points in the air between the sensor and the surface, which shows as noise close to the scanner and can degrade the SLAM solution in heavy conditions. Scan a pit once the dust has settled rather than following a haul truck. Heat affects endurance more than accuracy, mostly through battery performance, so carry more charge on long summer walks than the runtime figure suggests.