RuView: See Through Walls Using Nothing But WiFi Signals

What if your WiFi router could watch you? Not with cameras — that would be creepy, expensive, and a privacy nightmare. I'm talking about really watching you. Through walls. In total darkness. While you sleep. Measuring your breathing, counting your heartbeats, tracking every step, every gesture, every fall — all without a single pixel of video, without wearing anything, without even knowing it's there.

Sounds like science fiction? It's not. It's physics. And it's already running on a $9 chip sitting on a breadboard somewhere right now.

For decades, developers have been shackled to cameras, LiDAR, and wearables for spatial intelligence. Each comes with brutal trade-offs: cameras need light and line-of-sight, violate privacy regulations, and cost hundreds per zone. LiDAR can't see through walls. Wearables require user compliance — good luck getting grandma to charge her fall-detection pendant every night. The result? Gaps. Blind spots. Dead zones where people get hurt, where energy gets wasted, where businesses fly blind.

Enter RuView. This open-source platform transforms commodity WiFi signals into a complete spatial intelligence system — real-time human pose estimation, vital sign monitoring, presence detection, and environmental mapping — using nothing but the radio waves already flooding your space. Built on ESP32 hardware as cheap as a sandwich, running entirely on the edge with zero cloud dependency, RuView represents a fundamental shift in how we think about sensing. No cameras. No wearables. No internet required. Just pure signal processing wizardry turning invisible radio disturbances into actionable intelligence.

If you're building smart spaces, healthcare monitoring, security systems, or robotics, what you're about to read could change your entire architecture.

What is RuView?

RuView is an open-source WiFi sensing platform created by ruvnet that extracts Channel State Information (CSI) from standard WiFi signals and reconstructs rich spatial data about human bodies, activities, and environments. The name plays on "review" — a second look at your space through an entirely different lens — and "Ru" from its creator's handle, with the Greek letter π prefixing it as a mathematical constant, suggesting fundamental transformation.

The project builds directly on groundbreaking research from Carnegie Mellon University's "DensePose From WiFi" paper, which first demonstrated that WiFi signals contain sufficient information to estimate dense human correspondences — essentially mapping a 3D body surface from radio wave disturbances. RuView takes this academic proof-of-concept and engineers it into a production-ready platform with real-time performance, edge deployment, and practical applications spanning healthcare to robotics.

At its core, RuView treats every WiFi router as an illuminator and every ESP32 sensor as a receiver in a massive distributed radar system. When a person breathes, their chest moves approximately 4-12 millimeters — enough to shift the phase of 2.4GHz radio waves measurably. When they walk, their limbs create characteristic Doppler signatures. When they fall, the sudden change in body orientation produces a distinct scattering pattern. RuView's signal processing pipeline captures these micro-disturbances and reconstructs them into 17 COCO keypoint poses, breathing rates from 6-30 BPM, heart rates from 40-120 BPM, and presence detection with reported 100% accuracy at 0.012ms latency.

The platform is trending now because it solves three previously incompatible requirements simultaneously: privacy preservation (no cameras means no GDPR/HIPAA imaging compliance), penetration capability (WiFi passes through walls where optical sensors fail), and cost efficiency ($9 per ESP32-S3 node versus $200-2000 per camera zone). For developers building in healthcare, smart buildings, retail, and robotics, this triad is irresistible.

RuView's architecture is deliberately modular. The RuVector backbone handles attention mechanisms, graph algorithms, and compression. The Cognitum Seed provides persistent vector storage, kNN search, and cryptographic attestation via Ed25519 witness chains. And 65 WebAssembly edge modules — from sleep apnea detection to forklift proximity alerts — run directly on the ESP32 without cloud connectivity.

Key Features That Make RuView Insane

RuView isn't a single trick pony. It's a complete sensing stack with capabilities that sound made up until you understand the physics. Here's what separates it from anything else in the open-source sensing space:

🦴 Camera-Free Pose Estimation (17 COCO Keypoints) — Using the WiFlow architecture and 10-signal sensor fusion, RuView reconstructs full body poses without ever capturing an image. The current PCK@20 accuracy sits around 2.5% with proxy labels, but the camera-supervised training pipeline (ADR-079) targets 35%+ — and the infrastructure is already built, awaiting final data collection and evaluation. At 171K embeddings per second on an M4 Pro, this is real-time performance.

🫁 Contactless Vital Signs — Breathing detection via 0.1-0.5 Hz bandpass filtering with zero-crossing BPM calculation. Heart rate from 0.8-2.0 Hz bandpass. Both work through clothing, blankets, even thin walls. The medical edge modules include sleep apnea screening, cardiac arrhythmia detection, respiratory distress alerts, and seizure recognition via 6-state machines.

🧱 Through-Wall Penetration — Using Fresnel zone geometry and multipath modeling, RuView achieves up to 5 meters of sensing depth through standard construction materials. Concrete, drywall, shelving — the radio waves don't care. This enables applications in search-and-rescue, security, and smart home automation where line-of-sight is impossible.

🧠 On-Device Edge Intelligence — The entire model fits in 55 KB of memory on an ESP32. Spiking neural networks adapt to new environments in under 30 seconds. Multi-frequency mesh scanning across 6 WiFi channels uses neighboring access points as free radar illuminators, tripling effective sensing bandwidth without additional hardware.

🔐 Cryptographic Attestation — Every measurement is signed in an Ed25519 witness chain. This isn't just sensing; it's verifiable sensing. Critical for medical, legal, and security applications where data integrity matters.

🌐 Multi-Modal Fusion — Optional integration with camera depth (MiDaS), mmWave radar, and WiFi CSI into unified 3D point clouds — 19,000+ points per frame at 22ms pipeline latency. When you need maximum fidelity, RuView doesn't force you to choose between modalities.

📡 65 Production WASM Edge Modules — From quantum-inspired coherence detection to GOAP autonomous planning, from hyperbolic space embeddings to psycho-symbolic reasoning — these 5-30 KB modules deploy over-the-air and execute in under 10ms. All no_std Rust, all tested (609 tests passing), all running without cloud connectivity.

Use Cases: Where RuView Absolutely Dominates

The versatility of WiFi sensing creates application domains that simply don't exist for camera-based systems. Here are four concrete scenarios where RuView changes the game:

🏥 Elderly Care & Assisted Living (No Wearable Compliance Required)

Fall detection is the holy grail of eldercare technology, but wearables fail because people don't wear them. Cameras fail because they're invasive, require light, and create regulatory nightmares. RuView deploys a single $8 ESP32-S3 per room and provides continuous fall detection with sub-2-second alert latency, nighttime activity monitoring, and breathing rate tracking during sleep — all without any device on the person's body. The Sleep Apnea and Gait Analysis edge modules run locally, preserving dignity while ensuring safety. For facilities with 50 rooms, that's $400 in hardware versus $10,000+ for camera infrastructure — and zero ongoing cloud fees.

🏢 Office Space Utilization (Existing Infrastructure, Instant Insight)

Enterprise real estate teams spend millions guessing which desks and meeting rooms actually get used. RuView leverages existing enterprise WiFi access points to deliver presence latency under 1 second, meeting room no-show detection, and HVAC optimization based on real occupancy rather than scheduled bookings. The Energy Audit edge module tracks after-hours usage patterns; the HVAC Presence module implements departure countdowns to prevent heating empty spaces. A 500-person office typically sees 15-30% HVAC savings — often $50,000+ annually — with software-only deployment on hardware already installed.

🤖 Cobot Safety in Manufacturing (Through-Obstruction Detection)

Collaborative robots need to know when humans enter their workspace, but manufacturing environments are obstacle-rich: shelving, equipment, partial walls. LiDAR occludes. Cameras fail in dust and poor lighting. RuView's ESP32 mesh creates invisible safety zones that detect human presence even behind obstructions, with sub-100ms latency for emergency stops. The Forklift Proximity and Perimeter Breach modules provide layered safety without wiring new sensors throughout the facility. For automotive and electronics manufacturers already using ESP32 for IoT, this is a firmware update away from comprehensive human-robot collaboration safety.

🔥 Search & Rescue Through Rubble (WiFi-Mat Disaster Module)

When buildings collapse, finding survivors quickly determines outcomes. Traditional methods — acoustic sensors, thermal imaging, search dogs — have severe limitations. RuView's WiFi-Mat module (ADR-001) detects breathing signatures through 30cm of concrete, performs START triage color classification, and provides 3D localization — all with a portable ESP32 mesh and laptop that fits in a backpack. Firefighters can locate occupants through smoke and walls before entering hazardous structures. This isn't theoretical: the physics of 2.4GHz penetration through debris is well-established, and RuView's respiratory-focused signal processing isolates the 0.1-0.5 Hz breathing signature from environmental noise.

Step-by-Step Installation & Setup Guide

RuView offers three deployment tiers depending on your hardware commitment. Here's how to get each running:

Option 1: Docker Evaluation (No Hardware Required)

The fastest way to explore RuView's signal processing pipeline. Runs with simulated CSI data, perfect for algorithm evaluation and frontend development.

# Pull the multi-arch image (amd64 + arm64)
docker pull ruvnet/wifi-densepose:latest

# Run with web interface on port 3000
docker run -p 3000:3000 ruvnet/wifi-densepose:latest

# Open browser to explore the live observatory
curl http://localhost:3000

This gives you the full visualization stack — pose skeletons, vital sign graphs, room heatmaps — using deterministic reference signals. No ESP32 needed, no WiFi configuration, instant gratification.

Option 2: Live Sensing with ESP32-S3 ($9 Hardware)

For real-world deployment, you'll need CSI-capable hardware. The ESP32-S3 is the sweet spot — dual-core, WiFi 4, CSI extraction support, and absurdly cheap.

# Step 1: Flash the firmware using esptool
# Replace COM9 with your actual serial port (/dev/ttyUSB0 on Linux, /dev/cu.usbserial-* on macOS)
python -m esptool --chip esp32s3 --port COM9 --baud 460800 \
  write_flash 0x0 bootloader.bin 0x8000 partition-table.bin \
  0xf000 ota_data_initial.bin 0x20000 esp32-csi-node.bin

# Step 2: Provision WiFi credentials and target sink IP
python firmware/esp32-csi-node/provision.py --port COM9 \
  --ssid "YourWiFi" --password "secret" --target-ip 192.168.1.20

Critical requirements: ESP32-C3 and original ESP32 are not supported — they're single-core and lack the DSP horsepower for real-time CSI processing. The ESP32-S3's dual-core Xtensa LX7 handles the Fast Fourier Transforms and matrix operations that make this possible.

Option 3: Full System with Cognitum Seed ($140 Total BOM)

The recommended production deployment adds persistent memory, cryptographic attestation, and AI integration via the Cognitum Seed hardware.

# Start the RF room scan for environmental fingerprinting
node scripts/rf-scan.js --port 5006

# Launch the spiking neural network for real-time adaptive learning
node scripts/snn-csi-processor.js --port 5006

# Run the mincut algorithm for accurate person counting in multi-person scenarios
node scripts/mincut-person-counter.js --port 5006

The Seed receives CSI streams from ESP32 nodes, stores them in its persistent vector database (RVF format), performs kNN similarity search for anomaly detection, and maintains the Ed25519 witness chain for cryptographic attestation of all measurements.

No hardware yet? Verify the entire signal processing chain with the deterministic reference:

python archive/v1/data/proof/verify.py

This runs a known CSI signal through every processing stage and validates outputs against pre-computed ground truth — essential for CI/CD and regression testing.

REAL Code Examples from the Repository

Let's examine actual code from RuView's implementation, with detailed explanations of what's happening at each stage.

Example 1: Self-Supervised Pretraining (No Labels Needed)

This is where RuView's magic begins — learning from raw WiFi data without any human annotation:

# Step 1: Pretrain on raw CSI data — the model teaches itself what "normal" looks like
cargo run -p wifi-densepose-sensing-server -- --pretrain \
  --dataset data/csi/ \
  --pretrain-epochs 50

# Step 2: Fine-tune with pose labels for full 17-keypoint capability
cargo run -p wifi-densepose-sensing-server -- --train \
  --dataset data/mmfi/ \
  --epochs 100 \
  --save-rvf model.rvf

# Step 3: Extract environment fingerprints from live WiFi streams
cargo run -p wifi-densepose-sensing-server -- --model model.rvf --embed

# Step 4: Build searchable index for anomaly detection and room identification
cargo run -p wifi-densepose-sensing-server -- --model model.rvf --build-index env

What's happening here? The --pretrain flag activates contrastive learning: the model sees two segments of WiFi data and learns to predict whether they came from the same room, same person, or same activity. This creates a 128-dimensional embedding space where similar situations cluster together. The transformer backbone (~28K parameters) and graph neural network process 56 CSI subcarriers, outputting both environment fingerprints (for search/identification) and body poses (for tracking). The --save-rvf exports to RuVector Format — a compact binary with cryptographic signatures. The --build-index env creates a searchable vector database for real-time anomaly detection: if a new fingerprint doesn't match any known cluster, something unusual is happening.

Example 2: ESP32 Firmware Flashing and Provisioning

The bridge between raw hardware and intelligent sensing:

# Flash bootloader, partition table, OTA data, and application firmware
python -m esptool --chip esp32s3 --port COM9 --baud 460800 \
  write_flash \
    0x0       bootloader.bin \
    0x8000    partition-table.bin \
    0xf000    ota_data_initial.bin \
    0x20000   esp32-csi-node.bin

# Configure network parameters and sink destination
python firmware/esp32-csi-node/provision.py \
  --port COM9 \
  --ssid "YourWiFi" \
  --password "secret" \
  --target-ip 192.168.1.20

Memory layout explained: The ESP32-S3 has 4MB+ flash organized in partitions. 0x0 holds the bootloader that initializes hardware and selects which app partition to run. 0x8000 contains the partition table — a directory of what's where in flash. 0xf000 stores OTA (Over-The-Air) update metadata, enabling remote firmware upgrades. 0x20000 is the actual application — the CSI capture and streaming firmware. The provision.py script writes WiFi credentials to NVS (Non-Volatile Storage) and configures the UDP target where CSI packets stream. The --baud 460800 is critical: lower baud rates cause buffer overflows during high-frequency CSI capture.

Example 3: Real-Time Sensing Server with Multi-Modal Fusion

The production pipeline combining WiFi CSI with optional camera depth:

# Start the core sensing server with model loading
cargo run -p wifi-densepose-sensing-server -- \
  --model model.rvf \
  --embed \
  --fusion-mode wifi-camera \
  --depth-model midas

Architecture insight: The --fusion-mode wifi-camera activates the dual-modal pose estimation demonstrated in the pose fusion demo. WiFi CSI provides through-wall, privacy-preserving coarse pose; camera depth (via MiDaS monocular depth estimation) provides fine detail where line-of-sight exists. The system attention-weights each modality based on signal quality — when someone walks behind a wall, camera confidence drops and WiFi confidence dominates automatically. The --depth-model midas specifies the Intel MiDaS v3 model for monocular depth estimation, producing 19,000+ point clouds at 22ms pipeline latency.

Example 4: Claude Code Plugin for Complete Workflow Automation

RuView's developer experience extends to AI-assisted setup:

# Add RuView as a plugin marketplace source
claude /plugin marketplace add ruvnet/RuView

# Install the ruview plugin
claude /plugin install ruview@ruview

# Or try without installing from local clone
claude --plugin-dir ./plugins/ruview

# Available commands:
# /ruview-start      → Interactive onboarding (Docker demo / build from source / live ESP32)
# /ruview-flash      → Build firmware and flash to ESP32
# /ruview-provision  → Configure WiFi, channel, MAC filtering, mesh topology
# /ruview-app        → Launch sensing application (presence / vitals / pose / sleep / MAT / point cloud)
# /ruview-train      → Train models with optional GCloud GPU acceleration
# /ruview-advanced   → Multistatic tomography, cross-viewpoint fusion, mesh security
# /ruview-verify     → Run full test suite + deterministic proof + witness bundle generation

Why this matters: The plugin wraps 96 Architecture Decision Records, 8 domain models, and complex hardware interactions into conversational commands. New team members go from zero to live sensing in minutes rather than days. The /ruview-verify command is particularly powerful — it runs 1,463 tests, validates deterministic signal processing proofs, and generates cryptographically signed witness bundles for compliance documentation.

Advanced Usage & Best Practices

Mesh Topology Optimization: Single ESP32 deployments have limited spatial resolution. Deploy 2+ nodes in multistatic configuration for N×(N-1) radio links, dramatically improving pose accuracy through cross-viewpoint attention weighting. The optimal geometry places nodes at room corners, 2-4 meters apart, at varying heights.

Channel Hopping Strategy: RuView's multi-frequency mesh scans 6 WiFi channels, using neighbor access points as free illuminators. Configure your mesh to avoid your primary data channel during high-traffic periods — the ESP32 can hop channels in <2ms without losing CSI synchronization.

Coherence Gate Tuning: The sig_coherence_gate module uses z-score phasor gating with hysteresis (Accept/PredictOnly/Reject/Recalibrate states). For stable environments, tighten the z-threshold to reduce false positives. For dynamic spaces (retail, events), loosen it and rely more on temporal pattern modules.

Memory-Constrained Deployment: The full model is 55KB, but you can deploy subsets. Presence detection alone fits in 12KB. Vital signs in 18KB. Use cargo build --no-default-features --features presence-only for ultra-constrained deployments.

Witness Chain Integration: For medical and legal applications, enable Ed25519 signing on every measurement. The witness chain provides non-repudiable timestamps and cryptographic proof that data wasn't tampered with — essential for clinical trials and insurance documentation.

Comparison with Alternatives

The verdict: RuView uniquely combines privacy, penetration, pose estimation, and vital signs at commodity cost. Cameras win on pose accuracy but lose everywhere else. Radar matches privacy but lacks pose and costs more. LiDAR is precision overkill that can't see through walls. PIR is cheap but dumb — binary presence only. For applications requiring any two of {through-wall, privacy, pose, vitals, low cost}, RuView is the only option.

FAQ

Q: Is this legal? Can anyone just spy on people through walls? A: RuView senses using WiFi signals you control — your own routers and deployed ESP32 nodes. It cannot intercept or decode other networks' data (that's cryptographically impossible with WPA3). Legal deployment requires the same considerations as any sensor: notice, consent in private spaces, and compliance with local surveillance laws. The privacy advantage is that no images are ever captured, eliminating GDPR video regulation and HIPAA imaging compliance.

Q: How accurate is the pose estimation really? A: Currently ~2.5% PCK@20 with proxy labels — sufficient for activity classification and coarse tracking. The camera-supervised pipeline (ADR-079) targets 35%+ and is fully implemented; pending data collection and evaluation phases. For comparison, early Kinect achieved ~40% on similar metrics. The 17-keypoint output matches COCO format, compatible with existing pose analysis tools.

Q: Will this work with my existing WiFi router? A: For RSSI-only presence and motion detection: yes, any router. For full CSI capabilities (pose, vitals, through-wall): you need CSI-extraction capable hardware — ESP32-S3 ($9), Intel 5300, or Atheros AR9580. Most consumer routers don't expose raw CSI. The ESP32 mesh is the recommended approach.

Q: What's the catch with the $9 price? A: That's the ESP32-S3 module cost in volume. Total deployment for a room: 1-3 ESP32s ($9-27), USB power supplies ($5), and optionally a Cognitum Seed ($140) for persistent storage and AI. Compare to $200-2000 per camera zone plus NVR, cabling, and cloud storage. The real "catch" is that single-node spatial resolution is limited — you need 2+ nodes or a Seed for best results.

Q: How does this compare to the original CMU DensePose From WiFi research? A: RuView is a complete re-implementation and extension. The CMU work was research code with limited real-time performance. RuView adds: production ESP32 firmware, edge deployment, vital signs, 65 WASM modules, cryptographic attestation, multi-modal fusion, self-learning adaptation, and extensive documentation. It's the engineering that makes the science practical.

Q: Can I use this for medical diagnosis? A: RuView's medical edge modules provide screening and monitoring, not diagnosis. Sleep apnea detection flags potential events for clinical follow-up. Cardiac arrhythmia alerts suggest checking with a physician. The system is designed as a triage and continuous monitoring tool, not a replacement for medical-grade equipment. Always consult healthcare professionals for diagnostic decisions.

Q: What happens when WiFi standards evolve to WiFi 6/7? A: Higher frequencies (6GHz) and wider channels (320MHz) actually improve spatial resolution for CSI-based sensing. RuView's architecture is frequency-agnostic — the signal processing pipeline adapts to available subcarriers. The ESP32-S3 is WiFi 4, but the framework supports research NICs with WiFi 6 CSI extraction. Future ESP32 variants will extend capabilities natively.

Conclusion

RuView represents something rare in technology: a genuine paradigm shift that doesn't require waiting for new hardware. The WiFi signals are already there. The ESP32s are already $9. The physics of human bodies disturbing radio waves hasn't changed since WiFi was invented — we just finally built the signal processing and machine learning to interpret those disturbances with precision.

For developers, this is an invitation to rethink sensing architecture entirely. Why deploy cameras when radio works through walls? Why force wearables on users when contactless monitoring is possible? Why send data to clouds when edge intelligence fits in 55KB?

The platform is beta software — APIs and firmware evolve, the camera-supervised training pipeline awaits final evaluation, and community contributions are actively sought. But the foundation is solid: 1,463 tests passing, 65 edge modules implemented, real-time demos running, and a clear roadmap from research to production.

If you're building in smart spaces, healthcare, robotics, security, or anywhere spatial intelligence matters, RuView demands your attention. Start with the Docker demo, flash an ESP32-S3, or dive into the full repository. The future of sensing is invisible — and it's already in the air around you.

→ Explore RuView on GitHub