# Spatial Digital Twin

> An editable 3D planning prototype for comparing room layouts, checking clearances, and simulating queues before a space is rearranged.

Explore a layout before moving a single chair.

- URL: https://weaveward.com/work/spatial-digital-twin
- Stage: Prototype
- Year: 2026
- Industry: Hospitality and retail
- Context: Internal product
- Capabilities: Computer vision, Web apps, Mobile apps, AI solutions
- Technologies: Python, FastAPI, PostgreSQL, OpenCV, ONNX Runtime, OR-Tools, SimPy, React, three.js, Flutter

## The challenge

Rearranging a cafe, shop or warehouse means more than fitting furniture into a floor plan. A team needs to compare access, walking routes, seating, queues and service capacity before committing to the move.

Measurements also vary in quality. A planning tool should show where a dimension comes from and which parts still need checking.

## What we built

Spatial Digital Twin brings capture, layout editing and operational simulation into one planning workspace.

- **Capture and review.** The pipeline accepts capture bundles, video and photo sets. Reconstruction methods fall back according to available depth, poses and compute, while the quality report records the method and confidence
- **A 2D and 3D editor.** Teams can place furniture, adjust objects and compare layout options with versions, comments and review workflows
- **Clearance and route checks.** Geometry checks identify collisions, aisle widths, door and exit clearance, wheelchair routes and capacity constraints. Issues appear at their location in the layout
- **What-if scenarios.** Cafe service, retail checkout, event arrivals and warehouse picking can be simulated across multiple random seeds. Results show ranges rather than a single unexplained estimate
- **Suggested layouts.** A constraint solver proposes options for a goal such as seating capacity. Natural-language edits in English or Indonesian preview changes before the user saves them
- **An optional live layer.** Camera integrations map anonymous positions onto the twin for heatmaps, replay and movement patterns

## How it works

### A starting point for review

The reconstruction is a starting point for review. Dimensions carry confidence and verification flags; measurements with weak evidence need a reference scale and human checking.

### Numbers from geometry and simulation

Layout metrics and simulation values come from geometry and simulation code. Optional language-model features help interpret commands or summarise outputs; they do not invent room sizes or operational numbers.

## Current stage

This is an internal prototype tested with synthetic scans and demo layouts. Synthetic LiDAR checks compare room dimensions to known ground truth within 2%, while the documented cafe simulation runs a three-hour scenario across ten seeds in under five seconds.

Those results do not establish surveying accuracy for arbitrary phone footage. Native ARKit/ARCore capture and GPU reconstruction paths have not been validated end to end on real devices in the documented runs. Layout checks also do not replace professional safety, accessibility or architectural review.

## Where this fits

Retailers, hospitality teams and venue planners can compare changes before moving the physical space. Explore our [computer vision](/services/computer-vision), [web applications](/services/web-apps) and [mobile applications](/services/mobile-apps), or [discuss your planning workflow](/contact).