Digital visualization showing spatial computing for industrial space mapping with raw 360 video data transforming into interactive AR interface on a tablet.

XGRIDS to VPS: Turn a Lixel Capture into a Machine-Readable Localization Map

One walk gives you the splat people look at and the map machines query.

Sub-5 cm median localization accuracy. Deploy anywhere.

The output is a localization map, not only a reconstruction. Phones, headsets, and robots resolve their 6-DoF pose against it.
Request For Demo
Read Docs
Start Free

Export It. Upload It. Localize It.

MultiSet accepts XGRIDS point clouds and metric-scaled Gaussian splats as native VPS inputs. Export from a Lixel capture and upload it.

MultiSet reconstructs it into a 6-DoF localization map, with no separate photogrammetry pipeline in between.

Insta360 & MultiSet AI Visual Positioning System
Two exports, and both are valid inputs
XGRIDS units produce both a point cloud and a Gaussian splat from the same walk. Either can become the localization map. The splat path only works if the splat carries metric scale, which XGRIDS captures generally do because the hardware measures real distance rather than inferring it from images alone.

Supported exports: E57 point cloud, Gaussian splat
Capture hardware:
- Lixel K1: handheld, walked routes
- Lixel L2 Pro: larger sites, longer runs

From Lixel Capture to AR, Robotics, and Spatial AI in Three Steps

MultiSet 360 video to VPS: a handheld 360 camera scanning a warehouse aisle into a spatial point cloud
Export
Export either the point cloud or the Gaussian splat from the same capture. If you already produce the splat for a client-facing deliverable, that is the file, and no second export is needed.
MultiSet 360 video to VPS: raw 360 capture processing into a centimeter-accurate localization map
Upload
Log into the MultiSet Developer Portal. Select your input type, choose indoor or outdoor, and drop in the export. Scale and reconstruction are handled for you. Processing begins immediately.
MultiSet 360 video to VPS: phones, AR glasses and robots localize with 6-DoF pose on a warehouse map
Localize & Build
Your VPS map is live. Phones, headsets, robots, and spatial agents resolve 6-DoF pose against it via REST API. Scale across floors and buildings with MapSet, and keep versions and annotations current as the space changes.

Why XGRIDS Captures Suit the One-Capture Model

One Walk, Two Artifacts

The same capture yields a splat a client can walk through and a feature map a device localizes against. Most of the market gives you one or the other, and the second site visit is the expensive part on a live or secure site.
A Localization Map, Not Just a Model

A splat is something people look at. A VPS map is something devices query. Phones, headsets, and robots resolve their 6-DoF pose against it to sub-5 cm, across Unity, iOS, Android, WebXR, Meta Quest, and ROS 2.
Metric Scale Is the Deciding Property

A splat reconstructed from images alone is internally consistent and scale-free, and it renders perfectly either way. For localization that is fatal, because pose error grows with distance. Hardware that measures real distance carries scale through, which is why these captures work as inputs.
Walked Coverage, Walked Height

Handheld SLAM captures continuously along the route rather than from chosen stations, so it does not leave the gaps that cause a map to work in some places and fail in others, and it sits close to the height a device will later be held at.
Your Cloud or Ours

Run on MultiSet's public cloud, your private VPC, self-hosted on-prem, or fully on-device. Data is encrypted in transit and at rest, with enterprise identity and audit controls.
Campus-Scale with MapSet

Stitch XGRIDS maps from multiple floors and buildings into one MapSet, and mix them with your Matterport, NavVis, Leica, FARO and 360 scans in a single coordinate system. Users localize continuously as they move between spaces, with no manual transitions and no drift.

From the Same Lixel Capture: Digital Twin VPS Localization Map

XGRIDS turns a walk into a point cloud and a splat you view, measure and share. MultiSet turns the same walk into a localization map your devices and robots query in real time. Different output, different job, and many teams run both.
Pilot Project
CAPABILITY
A 3D model / digital twin
A MultiSet VPS map
What it produces
A viewable 3D model, mesh, or digital twin of the space
A 6-DoF localization map devices query for their exact position
Primary job
Document, visualize, measure, and plan the space
Position phones, headsets, robots, and agents in the space, live
Who consumes it
People reviewing the space on a screen
Devices on site, resolving where they are to centimeters
On site
Revisited later in a browser or viewer
Localizes on device in under 1500 ms, on-prem or offline
Capture sources
Often tied to one capture type or platform
Scan-agnostic: XGRIDS, Matterport, NavVis, Leica, FARO, and 360 video in one map
Where it fits
Captures and documents reality
Makes that same reality queryable for AR, robotics, and spatial AI
Frequently asked questions
Can you turn an XGRIDS Lixel capture into a VPS map?

Yes. Export either the point cloud or the Gaussian splat and upload it. MultiSet builds a 6-DoF localization map that apps and robots query for their exact position, rather than a reconstruction to look at.

Should I upload the point cloud or the Gaussian splat?

Either works, so use whichever your workflow already produces. If you export the splat for a client-facing deliverable, use that and skip the second export. The only requirement on the splat path is metric scale.

How do I know if my Gaussian splat is metric-scaled?

Measure something you know inside it, a door height or a pallet width, and see whether it comes back correct. A splat reconstructed from images alone is internally consistent but scale-free, and it renders perfectly either way. Hardware that measures real distance carries scale through, which is why these captures generally work as inputs.

Why does metric scale matter so much?

Because a map that is not metric is a picture of a place. Pose returned in the wrong scale puts a label in the wrong position by an amount that grows with distance from the origin. For viewing it does not matter at all. For localization it is fatal.

Can I combine an XGRIDS map with my other scans?

Yes. MapSet stitches maps from different capture types into one coordinate system. Mix XGRIDS, Matterport, NavVis, Leica, FARO and 360 video across floors and buildings, and devices localize continuously across all of them.

How accurate is VPS from a Lixel capture?

Sub-5 cm median, with pose resolved in under 1500 ms. Accuracy holds through changing light and people moving through the space, because matching is against the structure of the building rather than against what is sitting in it.

How large a site can one capture cover?

There is no ceiling on total coverage. Handheld SLAM is usually chosen because the site is too large or too cluttered for static scanning, and MapSet stitches multiple runs into one continuous map across floors and buildings.

Request a Demo

→

Start Building

→
Pages
HomePricingBlogGet In TouchPrivacy Policy
Platform
VPSObject Tracking3D MappingE57 to VPS3DGS to VPS360 to VPSCase StudyMap Versioning
Developers
TechnologyPartnersDocumentationVPS ApplicationsVuforia AlternativeASA AlternativeStatusCommunity
contact us
28 Geary Street
STE 650 Suite #371,
San Francisco,
California 94108, USA
Follow us
Hexagonal 3D geometric shape with purple triangular facets in varying shades.
© 2026 Multiset AI
All rights reserved
Terms of use
·
Privacy Policy