Point Cloud
A point cloud is a dense set of three-dimensional points that describe the shape of surfaces in a space. Each point has coordinates in a shared frame. Together they form a 3D model of aisles, racking, floors, vehicles, and other geometry the sensors can see. Plants meet point clouds in survey scans from LiDAR scanning or photogrammetry, in digital twin geometry projects, and in onboard SLAM stacks that use cameras or lidar.
On a live floor, camera-based and lidar-based SLAM both build or update point-cloud style maps while the vehicle drives. That map supports localization of the truck that carries the unit. The same geometry can also feed object recognition for people, other vehicles, and loads in view. Continuous shopfloor rescanning on a daily or hourly cadence, so the twin geometry stays as fresh as the live positions, is still mostly a near-future roadmap rather than a standard plant product today.
A point cloud is geometry. It is not a barcode, WMS ID, or MES order by itself. Presence and shape can be detected in view. Business identity still needs a scan, tag, or system link when you must know which cage or which part number is on the forks.
Key Components
3D points: Millions of measured locations that approximate walls, racks, floors, and objects.
Sensor source: Cameras (including stereo or depth) or lidar on a vehicle, robot, or survey kit that capture the cloud.
Coordinate frame: Alignment to the plant model so the cloud sits on the same map as live locations.
Update cadence: One-time survey, occasional remapping after a layout change, or future continuous rescans from fleet traffic.
Downstream use: Localization, obstacle awareness, object detection, and twin geometry refresh.
Applications in Manufacturing and Logistics
SLAM forklifts and AMRs use point-cloud style maps to hold about 10 cm (4 in) pose accuracy without a dense radio ceiling. The same view can flag people or other vehicles in range and, with some vendor add-ons, pallet presence on the forks. Survey teams also drop point clouds into twin projects when CAD is stale. The floor deep dive sits under SLAM for RTLS.
Logistics twins that already show live movers can later absorb fresher geometry from fleet-built clouds once continuous rescan products mature. Until then, treat map refresh as a planned job after rack moves, not as an automatic hourly twin update.
Benefits and Challenges
Benefit: rich 3D shape of the real hall, not only a 2D drawing. Supports precise vehicle pose and visual detection of nearby objects without tagging everything in view.
Challenge: files are large, alignment to the plant model takes care, and dynamic inventory changes the cloud every shift. Continuous daily or hourly rescan for twin geometry is not a standard offering yet. Lighting, glass, and occlusion still limit what cameras and lidar can see.
Related Terms
Point clouds sit next to LiDAR scanning and photogrammetry as the usual survey outputs, next to mesh geometry as what many twin viewers load after the cloud is surfaced, next to SLAM as the map many camera and lidar stacks build, and next to digital twin, BIM, and CAD as other sources of plant shape. Location context sits under RTLS and indoor positioning.
Frequently Asked Questions
Is a point cloud the same as a digital twin? No. A point cloud is 3D geometry. A twin adds live state, systems links, and ops views on top of a model that may come from a cloud, CAD, or both.
Do camera and lidar SLAM both create point clouds? Yes. Both families build dense 3D maps from what the vehicle sees. The sensor physics differ. The plant output is still a map plus a live pose.
Can we rescan the shopfloor every hour today? Continuous daily or hourly rescan from fleet traffic is a near-future roadmap for many stacks, not a widely deployed plant product yet. Plan map updates after layout change until that lands.
Does a point cloud identify every pallet? It can support presence and shape detection. Named inventory identity still needs WMS, a scan, a tag, or a dedicated presence add-on tied to business IDs.