Spatial Ground Truth for Pharmaceutical Manufacturing

A plant datum with a stated tolerance, built from scans you already own. Phones, wearables and autonomous robots all resolve against it, so a technician knows which asset they are facing, and every observation carries the position it was taken from.

Positioning for People, Robots and Automated Lines

One reference frame for the whole site
From warehouse to filling room.
One frame for technicians, wearables and autonomous robots.
Built from the scans you already own. Nothing installed on the floor.
Private cloud through to fully on-device, on the same API.
A pharmaceutical manufacturing suite with a single coordinate frame overlaid across the floor and markers on equipment at different depths
Measured on site

12 minutes to under 90 seconds. Average time to locate a single asset, measured during a proof of concept at a pharmaceutical manufacturer, across three sites. 87% faster asset location, 3,520 hours returned to productive work each year, and roughly 20 FTE-months given back to the business.

The Limits of Paper-Based Verification

Nobody made a mistake. The record says the work was done. It does not show it.
A long row of visually identical manual valves on overhead pipework, with only one of them marked. A route along a packing line passing ten checkpoint markers, only two of which are recorded. Three record panels each pointing at a different skid in a row of identical process skids.

Finding the asset is tribal knowledge

A plant can carry thousands of near-identical valves, many behind equipment or mounted high above the floor. A spreadsheet gives general directions. The technician who knows where things are is the system, and that knowledge leaves when they do.

An inspection that cannot be proved did not happen

Packing lines run dozens of checkpoints against paper procedures. Nothing records that a checkpoint was visited, in what order, or how thoroughly it was examined. The document says the work was done. It does not show it.

Every system models the plant. None locates the work.

The MES holds the batch, the CMMS holds the work order, the twin holds the geometry. Each is correct, and each assumes somebody already knows which asset is in front of them and where they were standing. That assumption is the gap.

Asset Finding, Line Clearance and Inspection Evidence

One map, and the record knows where it was made.
A technician with a tablet at a process skid, one valve outlined and a route line running along the floor to where they stand
Asset finding and maintenance routing
Route a technician to a specific valve among thousands, including assets hidden behind equipment or well above head height. A work order deep-links straight into the route, so the technician opens the job and walks. The record stays in the system that owns it. We hold the coordinate and the tag that points to it.
An operator at a packing line with four confirmed checkpoint markers along the machine and a position ring on the floor beneath them
Line clearance verification
Confirm that each checkpoint on a line was visited, and in what order. Position is recorded per observation, so the clearance record carries evidence of where the operator stood rather than a signature alone.
An operator reading a gauge on a process vessel, with a marker on the floor recording the position the reading was taken from
Inspection evidence with a position attached
Record the exact spot a reading or an image was captured from, so the next visit reproduces the same view and the comparison is like for like.

Object Tracking for Line Clearance and Isolation

Object Tracking recognizes a specific pump, valve or skid from its CAD or scan model and localizes against the asset rather than the room. For isolation checks, close-quarters work and confirming which of several identical units is in front of you.

Ingest

Build the map from capture you already own. E57 and LiDAR point clouds, Matterport, Leica, NavVis, Faro, XGRIDS, 3D Gaussian Splats, 360 video, CAD or BIM. Use the right instrument per zone and stitch them into one frame. Nothing is installed on the floor.

Tag

Tag assets from a phone on the floor or from the desktop dashboard. Detection proposes candidates and your subject matter expert approves them, so the tagging load falls on review rather than data entry.

Localize

A single camera frame returns position and orientation in the plant frame. Phones, wearables and autonomous robots resolve against the same map, so a mixed fleet sees the same thing in the same place.

Stay true

Scheduled re-checks hold the registration true through changeovers, shutdowns and moved equipment. Map versioning keeps a partial re-scan in the same frame, and every position stays traceable to the version it was captured against.
Two capture passes over a manufacturing suite merging into one continuous map. Assets being tagged from a phone on the floor, with three tags confirmed and one proposed. Three devices at different angles each resolving to one shared origin on the floor. A suite part way through a changeover, with anchors and the route holding their positions.

Scan-Agnostic Localization, Built as Infrastructure

Scan-agnostic
Build the map from E57, LiDAR, Matterport, 3DGS, 360 video, CAD or BIM. No proprietary scanner, no re-capture.
Two to three days per site
A site comes online in two to three days from scans it already holds. Nothing is mounted, cabled or calibrated on the floor.
Two primitives, one SDK
VPS for the environment, Object Tracking for the equipment. Same map, same SDK.
Built for changeovers
Relocalization holds through changeovers, shutdowns and moved equipment. A changed area is re-scanned into the same frame rather than recaptured whole.
Your map, your infrastructure
Public cloud, private cloud, self-hosted or fully on-device. In self-hosted and on-device modes, no maps, images or spatial queries leave your infrastructure.
Tagging your team can own
Tag assets from a phone on the floor or from the desktop dashboard. Detection proposes candidates and your subject matter expert approves them.
Phones, wearables and robots
Smart glasses, iOS, Android, Unity, WebXR, REST APIs and ROS 2 ship from the same map. The line that guides a technician today guides an autonomous robot on the same frame tomorrow.

Where Your Plant Scans Are Allowed to Live

An interior scan of a manufacturing site is sensitive before it is useful. Room layouts, equipment positions, how material moves, what changed at the last changeover. On a regulated site, where that data lives is a procurement question before it is a technical one.

MultiSet runs where your compliance team says it can live, from public cloud to a fully on-device deployment with no outbound call, on the same API.

It fits the frame you already run: plant grid or local survey coordinates with an EPSG code, georeferenced to WGS 84, so positions come back in the system your teams already use.
Diagram showing private cloud, on-prem and on-device deployment connecting to one MultiSet core that serves warehouse, factory, construction, healthcare and tunnel environments

Where This Sits Against Paper and Fixed Infrastructure

Paper and fixed infrastructureMultiSet positioning layer
Proof an inspection happenedA signatureThe position it was taken from
Finding a specific assetTribal knowledge and a spreadsheetRouted, to the asset
Bringing a site onlineA projectTwo to three days, nothing installed
Evidence over timeHard to compare between visitsSame position, same view, versioned
Where the data livesWherever the vendor puts itYour cloud, your estate, or the device

Stated plainly: we are not replacing your quality system. We are giving the record a place to stand.

Two related pages: Connected worker covers the field-of-view layer this sits on, and AR quality inspection covers verification and evidence capture inside the workflow.

Pharmaceutical VPS:
Frequently Asked Questions
How quickly can a site go live?

Two to three days per site, using scans the site already holds. Nothing is installed on the floor, so there is no mounting, cabling or calibration work and no capex tied to the building. Tagging assets is the variable, and it can start from a phone on the floor or from the desktop dashboard.

Do you need a new scan of our plant?

Usually not. E57, LiDAR point clouds, Matterport, Leica, NavVis, Faro, XGRIDS, 3D Gaussian Splats, 360 video, CAD and BIM all work as inputs. Most sites already hold something we can build from, and zones can be captured with different instruments and stitched into one frame.

Is this a validated system?

No, and it is not sold as one. MultiSet returns position and orientation, and records the position an observation was taken from. It is not a quality management system, not a batch record, and it does not make a site compliant. It supplies evidence that your validated systems can reference.

How accurate is it, and under what conditions?

Centimeter-scale in mapped areas with adequate visual texture. Accuracy depends on the quality of the source capture and how much the space has changed since it was taken, which is why scheduled re-checks and map versioning are part of the deployment rather than an afterthought.

Can it run with no network, or inside our own cloud?

Yes. Public cloud, private cloud, self-hosted, or fully on-device. In self-hosted and on-device modes, no maps, images or spatial queries leave your infrastructure, and no external connection is needed to return a position.

What happens when we change a line or move equipment?

Re-scan the zone that changed rather than the whole site. Anchored content carries forward to the new version, and every recorded position stays traceable to the map version it was captured against.

One Frame for Every Site in the Network