Wireless Integrated Gaussian Live Splatting
Reality,
reconstructed.
Every crowded room is already full of radio. WIGLS reads the WiFi and RF signals bouncing off people and structure, measures motion with Doppler analysis, and fuses it with Gaussian splatting into a digital twin that stays live.
Digital twins are slow, manual, and expensive.
- LiDAR scans need specialist operators and a closed site.
- Photogrammetry needs a camera crew and hours of processing.
- By the time a twin is usable, the space has changed.
- Live events vanish the moment they end.
- Physical spaces stay disconnected from the systems that run them.
Build them from the signal already in the air.
Wireless signals scatter off every body and wall in a venue. That scatter carries geometry and motion. WIGLS captures it with a mesh of edge nodes, turns it into velocity and occupancy with Doppler analysis, and fuses it with cameras and Gaussian splatting into one continuously updating twin that replays on demand and streams anywhere.
We start where the technology is proven today, camera-first Gaussian capture with wireless sensing filling the gaps, and move toward reconstruction from ambient signal as the science matures. The science section is explicit about which is which.
What WiFi Doppler actually measures.
A WiFi receiver sees the transmitted wave plus every copy of it that reflected off the room. When a reflector moves, its copy arrives with a slightly shifted frequency. That shift is the Doppler frequency, and it is proportional to how fast the reflected path is getting shorter or longer.
- fDDoppler frequency shift of the reflected path, in hertz. Positive when the path is shortening (blue-shift), negative when it lengthens (red-shift).
- vtx, vrxComponents of the reflector's velocity along the lines to the transmitter and the receiver. For a body moving straight at a co-located transmitter and receiver this sums to twice the walking speed.
- fcCarrier frequency: 2.4, 5 or 6 GHz for WiFi. Higher carrier, larger shift, finer velocity resolution.
- cSpeed of light. The wavelength at 5.2 GHz is 5.8 cm, so every 5.8 cm of path change rotates the received phase by a full cycle.
In practice a walking person produces a spread of shifts, not one line: the torso moves at one speed, arms and legs swing faster, and each limb's echo arrives from a different angle. That spread over time is the Doppler spectrogram, the same readout under the hero above. The Widar family of systems from Tsinghua showed that with several receivers you can convert those spectrograms into a body-velocity profile that is independent of where the person stands, which is what makes gesture and activity recognition work across rooms.
Doppler calculator
Change the speed and angle to see the shift a WiFi node would observe.
Assumes a co-located transmitter and receiver, so the path-length rate is twice the radial speed. Real deployments sample channel state at 200 to 1,000 Hz to leave headroom for limb motion.
Where the signal comes from: channel state information.
Every WiFi frame is sent across dozens to hundreds of narrow subcarriers. The receiver estimates the amplitude and phase of each one to decode the data, and that estimate, the channel state information (CSI), is a fingerprint of the room at that instant. Record it several hundred times a second on a few antennas and you have a movie of how the radio environment is changing.
Consumer chips corrupt the phase with carrier and sampling offsets. The standard fix is to multiply one antenna's CSI by the conjugate of another's so the shared offsets cancel, leaving the Doppler of the moving reflectors intact. Amplitude alone is enough for coarse jobs like presence; phase is what unlocks velocity.
What wireless sensing can do today, and what it cannot yet.
This is the ladder WIGLS climbs. Each rung is either shipping in commercial products, demonstrated in peer-reviewed research, or still open. The open rungs are the company's research agenda, not its promise for a first deployment.
- Presence and motionIs anyone in the room, and are they moving? Shipping in consumer routers and ISP mesh systems as "WiFi motion".Shipping
- Breathing and heart rateMillimetre chest displacement read from phase: respiration to within 0.2–0.7 breaths per minute for one still person within a few metres, up to four people with multi-link systems. Commercial in sleep and elder-care devices.Shipping
- Fall detectionThe Doppler signature of a fall is sharp and distinctive: 90–94% precision with 13–15% false alarms in the WiFall studies. Commercial in home-care products.Shipping
- Gesture and activityWidar 3.0 reaches 92.7% on gestures in a known room and 82–92% across new rooms, positions and users, on commodity hardware. One person at a time.Research, mature
- Localization and trackingTracking to tens of centimetres for one or two people, using phase-based angle, time and Doppler estimates across several links. Two walkers is the practical limit per link.Research, mature
- Crowd counting and densityEstimating how many people are in an area from how much the channel churns. A 2020 study counted 34 people in one room to within one person 97.7% of the time. Nothing published works at crowd scale yet.Research, early
- Human poseDensePose From WiFi (Carnegie Mellon, 2023) recovered body surface and 17 keypoints from a 3×3 link, but only in the rooms it was trained in and it fails with three or more people. Mesh systems reach 2–5 cm joint error under the same limits.Research, early
- Static geometryImaging walls and furniture from WiFi works at wavelength scale (6–12 cm) and is slow. Radio-frequency Gaussian splatting (RF-3DGS, SwiftWRF) now models a venue's radio field in minutes, which is the map WIGLS places its nodes on.Open
- Photoreal reconstruction from signal aloneNobody has done it. Radio carries no colour or texture, so cameras stay in the loop. This is the far end of the WIGLS roadmap, not the start.Open
The hardware is cheap and the standard is arriving.
Channel state information used to require a decade-old Intel card and a patched Linux driver. Today a five-dollar microcontroller exports it, a Raspberry Pi can be patched to do it on 80 MHz channels, and IEEE 802.11bf, the WLAN Sensing amendment published in September 2025, defines how access points and clients request and report sensing measurements so the next generation of WiFi 7 and 8 chips does it natively. Qualcomm announced its first 802.11bf-capable platforms in March 2026.
A WIGLS edge node is a small computer with two or three WiFi radios, an optional camera, and enough compute to run the Doppler pipeline locally. Only velocity fields, occupancy and, where a camera is present, encoded splat updates leave the node. Raw channel data never does: research has shown a person can be re-identified from their WiFi signature, so WIGLS treats radio like a biometric and outputs aggregates only.
| Platform | What it gives you | Cost per node |
|---|---|---|
| ESP32 (ESP-CSI) | 1 antenna, 20 MHz, CSI at hundreds of frames per second. Enough for presence, breathing, coarse motion. | ~$5 |
| Raspberry Pi 4 + Nexmon CSI | Broadcom radio patched to export CSI on 80 MHz channels with up to 4 spatial streams. | ~$60 |
| Intel AX210 + PicoScenes | Modern WiFi 6E card, 160 MHz channels, 2.4/5/6 GHz, high-rate CSI on a Linux mini-PC. The research workhorse. | ~$150–300 |
| Software-defined radio | Full control of waveform and timing for research on multi-band and bistatic geometries. | $1,000+ |
| 802.11bf-capable chipsets | Standardised sensing measurement setup and CSI reporting between any compliant devices. | Built in |
Reading list
The work WIGLS builds on. Start with the survey if you are new to the field.
- WiFi Sensing with Channel State Information: A SurveyMa, Zhou, Wang. ACM Computing Surveys, 2019. The standard map of the field.
- WiSee: Whole-Home Gesture Recognition Using Wireless SignalsPu, Gupta, Gollakota, Patel. MobiCom 2013. The paper that showed Doppler gestures on WiFi.
- Understanding and Modeling of WiFi Signal Based Human Activity Recognition (CARM)Wang et al. MobiCom 2015. The CSI-speed model linking channel change to body velocity.
- Widar 3.0: Zero-Effort Cross-Domain Gesture Recognition with WiFiZheng et al. MobiSys 2019, IEEE TPAMI 2021. Body-velocity profiles from multi-receiver Doppler; public dataset.
- DensePose From WiFiGeng, Huang, De la Torre. Carnegie Mellon, 2023. Human surface and pose from a 3×3 WiFi link.
- NeRF²: Neural Radio-Frequency Radiance FieldsZhao et al. MobiCom 2023. Neural fields that model how radio propagates through a scene.
- RF-3DGS: Wireless Channel Modeling with Radio Radiance Field and 3D Gaussian Splatting2024–2025. Gaussian splats of the radio field: minutes to train, milliseconds to render.
- SwiftWRF: Wireless Radiance Fields with Deformable 2D Gaussians2025–2026. Radio-field rendering above 100,000 frames per second.
- IEEE 802.11bf: Enabling the Widespread Adoption of Wi-Fi SensingOverview of the WLAN Sensing amendment: roles, sessions, trigger-based sounding and CSI reporting.
- IEEE Std 802.11bf-2025: WLAN SensingApproved May 2025, published September 2025. The standard itself.
- 3D Gaussian Splatting for Real-Time Radiance Field RenderingKerbl et al. SIGGRAPH 2023. The rendering primitive WIGLS twins are built from.
Five phases from capture that works today to signal-native world models.
Each phase ships something customers pay for and produces the paired signal-and-scene data the next phase is trained on.
Multi-camera Gaussian capture
Production-grade splat capture from drone and camera arrays. Shippable now with the founder's existing drone network; every event captured becomes training data.
WiFi-assisted reconstruction
Doppler velocity fields and occupancy from edge nodes fill occlusions, cut the camera count, and keep the twin moving between camera updates.
Wireless scene understanding
Models trained on phases 1 and 2 infer motion, occupancy and coarse geometry from signal alone inside instrumented venues.
Near-real-time twins
The edge mesh streams continuously updating twins with sub-second latency to broadcast, operations and VR.
Signal-native world models
A spatial foundation model that reconstructs a space directly from ambient wireless signal, with cameras optional.
One platform, five deployments.
WIGLS LiveReal-time event reconstruction
WIGLS VenuePermanent venue mapping
WIGLS CityUrban-scale digital twin
WIGLS IndustrialFactory and warehouse intelligence
WIGLS DefensePersistent situational awareness
A new category: wireless spatial intelligence.
Digital twin, XR, smart city, event tech, robotics and defense combined, by 2034.
Real-time spatial intelligence for events, venues, industry and public safety.
Targeted capture across the beachhead verticals below.
Why now
Competitive landscape
| Company | Focus | Live, real time | Wireless-native |
|---|---|---|---|
| Matterport | Static 3D scans | — | — |
| Niantic Spatial | AR maps and visual positioning | — | — |
| NVIDIA Omniverse | Simulation and authoring | — | — |
| Unity / Unreal | Game and render engines | — | — |
| Hexagon | LiDAR and survey | — | — |
| Bentley Systems | Infrastructure modelling | — | — |
| Origin AI (bought by ADT for $170M, Feb 2026), Cognitive Systems | WiFi motion sensing for homes and ISPs | ||
| WIGLS | Wireless spatial intelligence for venues, cities and defense |
Incumbents reconstruct space with cameras, LiDAR or simulation, captured once and authored by hand. The WiFi sensing companies detect motion but do not reconstruct anything. WIGLS sits between them: a signal-to-scene data engine and spatial foundation models that neither side is positioned to build.
Five-year projection
Revenue model
- 1 active twin
- Cloud replay
- Basic analytics
- Email support
- Multi-event twins
- VR delivery
- Crowd analytics
- API access
- Persistent twins
- Edge mesh deployment
- White label
- SLAs and SSO
- On-prem or air-gapped
- Denied-environment sensing
- Compliance and ATO path
- Dedicated support
Additional revenue from edge-node hardware, data licensing, API metering, analytics packages and white-label licensing.
Go to market
- Year 1Live eventsMusic festivals, concert promoters, venue operators.
- Year 2VenuesSports organisations and convention centres.
- Year 3IndustrialLogistics, manufacturing and energy facilities.
- Year 4Smart citiesMunicipal and transportation deployments.
- Year 5DefenseDefense, emergency management and government.
Pre-seed hires
- Founding ML / spatial AIGaussian splatting and neural rendering
- RF / wireless sensing leadCSI, Doppler, signal processing
- Edge systems engineerMesh nodes, real-time inference
- Full-stack / platformTwin delivery, API, dashboards
- Field and solutionsFestival and venue deployment
- PartnershipsPromoters, venues, government channels
To build the founding team, ship phase 1 and 2 capture, deploy the first edge-mesh pilots at live events, and stand up the signal-to-scene data engine.
Deck outline
- Vision: reality, reconstructed
- Problem: twins are slow, manual, expensive
- Solution: wireless spatial intelligence
- Product: Live, Venue, City, Industrial, Defense
- Technology and moat: signal-native world models
- Market: TAM, SAM, SOM
- Competition: a new category
- Go to market: events, then cities, then defense
- Traction and roadmap: five phases
- Team and founder story
- Financials: five-year projection
- The ask: $3M pre-seed
Market sizing and financial figures are illustrative management projections for planning and discussion, not guarantees of future performance. Technology phases describe a development roadmap; capabilities are delivered incrementally, and the science section states which capabilities are proven today.
Built from the field.
WIGLS was founded by a media producer with 18 years of global content creation behind him: festivals, live events and immersive storytelling, backed by a working drone network and long-standing venue relationships.
After watching countless events disappear the moment they ended, he set out to build a system that could preserve and reconstruct them, starting with the signals that already fill every space. WIGLS is the operating system for spatial intelligence.
Let's reconstruct your world.
Booking a venue, running an event, building a smart-city program, or investing in the category: tell us what you want to see and we will reply within two business days.
Or write directly: admin@thebaileyperspective.com