WiFi DensePose User Guide

WiFi DensePose turns commodity WiFi signals into real-time human pose estimation, vital sign monitoring, and presence detection. This guide walks you through installation, first run, API usage, hardware setup, and model training.

Table of Contents

1. Prerequisites
2. Installation
   • Docker (Recommended)
   • From Source (Rust)
   • From crates.io
   • From Source (Python)
   • Guided Installer
3. Quick Start
   • 30-Second Demo (Docker)
   • Verify the System Works
4. Data Sources
   • Simulated Mode (No Hardware)
   • Windows WiFi (RSSI Only)
   • ESP32-S3 (Full CSI)
   • ESP32 Multistatic Mesh (Advanced)
   • Connect Mesh Data to the Dashboard and Observatory
   • Cognitum Seed Integration (ADR-069)
5. REST API Reference
6. WebSocket Streaming
7. Web UI
8. Vital Sign Detection
9. CLI Reference
10. Observatory Visualization
11. Loading the Pretrained Model from Hugging Face
12. Adaptive Classifier
    • Recording Training Data
    • Training the Model
    • Using the Trained Model
13. Training a Model
    • CRV Signal-Line Protocol
14. RVF Model Containers
15. Hardware Setup
    • ESP32-S3 Mesh
    • Intel 5300 / Atheros NIC
16. Camera-Free Pose Training
17. ruvllm Training Pipeline
18. Docker Compose (Multi-Service)
19. Testing Firmware Without Hardware (QEMU)
    • What You Need
    • Your First Test Run
    • Understanding the Test Output
    • Testing Multiple Nodes at Once (Swarm)
    • Swarm Presets
    • Writing Your Own Swarm Config
    • Debugging Firmware in QEMU
    • Running the Full Test Suite
20. Troubleshooting
21. FAQ

Prerequisites

Hardware for live sensing (optional):

No hardware? The system runs in simulated mode with synthetic CSI data.

Installation

Docker (Recommended)

The fastest path. No toolchain installation needed.

docker pull ruvnet/wifi-densepose:latest

Multi-architecture image (amd64 + arm64). Works on Intel/AMD and Apple Silicon Macs. Contains the Rust sensing server, Three.js UI, and all signal processing.

Data source selection: Use the CSI_SOURCE environment variable to select the sensing mode:

Example: docker run -e CSI_SOURCE=esp32 -p 3000:3000 -p 5005:5005/udp ruvnet/wifi-densepose:latest

From Source (Rust)

On Debian/Ubuntu-based Linux systems, install the native desktop prerequisites before the first Rust release build:

sudo apt update
sudo apt install -y \
  build-essential pkg-config \
  libglib2.0-dev libgtk-3-dev \
  libsoup-3.0-dev \
  libjavascriptcoregtk-4.1-dev \
  libwebkit2gtk-4.1-dev

This prepares the native GTK/WebKit dependencies used by the desktop/Tauri crates in this workspace.

git clone https://github.com/ruvnet/RuView.git
cd RuView/v2

# Build
cargo build --release

# Verify (runs 1,400+ tests)
cargo test --workspace --no-default-features

The compiled binary is at target/release/sensing-server.

From crates.io (Individual Crates)

All 16 crates are published to crates.io at v0.3.0. Add individual crates to your own Rust project:

# Core types and traits
cargo add wifi-densepose-core

# Signal processing (includes RuvSense multistatic sensing)
cargo add wifi-densepose-signal

# Neural network inference
cargo add wifi-densepose-nn

# Mass Casualty Assessment Tool
cargo add wifi-densepose-mat

# ESP32 hardware + TDM protocol + QUIC transport
cargo add wifi-densepose-hardware

# RuVector integration (add --features crv for CRV signal-line protocol)
cargo add wifi-densepose-ruvector --features crv

# WebAssembly bindings
cargo add wifi-densepose-wasm

# WASM edge runtime (lightweight, for embedded/IoT)
cargo add wifi-densepose-wasm-edge

See the full crate list and dependency order in CLAUDE.md.

From Source (Python)

git clone https://github.com/ruvnet/RuView.git
cd RuView

pip install -r requirements.txt
pip install -e .

# Or via PyPI
pip install wifi-densepose
pip install wifi-densepose[gpu]   # GPU acceleration
pip install wifi-densepose[all]   # All optional deps

Guided Installer

An interactive installer that detects your hardware and recommends a profile:

git clone https://github.com/ruvnet/RuView.git
cd RuView
./install.sh

Available profiles: verify, python, rust, browser, iot, docker, field, full.

Non-interactive:

./install.sh --profile rust --yes

Quick Start

30-Second Demo (Docker)

# Pull and run
docker run -p 3000:3000 -p 3001:3001 ruvnet/wifi-densepose:latest

# Open the UI in your browser
# http://localhost:3000

You will see a Three.js visualization with:

• 3D body skeleton (17 COCO keypoints)
• Signal amplitude heatmap
• Phase plot
• Vital signs panel (breathing + heartbeat)

Verify the System Works

Open a second terminal and test the API:

# Health check
curl http://localhost:3000/health
# Expected: {"status":"ok","source":"simulated","clients":0}

# Latest sensing frame
curl http://localhost:3000/api/v1/sensing/latest

# Vital signs
curl http://localhost:3000/api/v1/vital-signs

# Pose estimation (17 COCO keypoints)
curl http://localhost:3000/api/v1/pose/current

# Server build info
curl http://localhost:3000/api/v1/info

All endpoints return JSON. In simulated mode, data is generated from a deterministic reference signal.

Data Sources

The --source flag controls where CSI data comes from.

Simulated Mode (No Hardware)

Default in Docker. Generates synthetic CSI data exercising the full pipeline.

# Docker
docker run -p 3000:3000 ruvnet/wifi-densepose:latest
# (--source auto is the default; falls back to simulate when no hardware detected)

# From source
./target/release/sensing-server --source simulate --http-port 3000 --ws-port 3001

Windows WiFi (RSSI Only)

Uses netsh wlan to capture RSSI from nearby access points. No special hardware needed. Supports presence detection, motion classification, and coarse breathing rate estimation. No pose estimation (requires CSI).

# From source (Windows only)
./target/release/sensing-server --source wifi --http-port 3000 --ws-port 3001 --tick-ms 500

# Docker (requires --network host on Windows)
docker run --network host ruvnet/wifi-densepose:latest --source wifi --tick-ms 500

Community verified: Tested on Windows 10 (10.0.26200) with Intel Wi-Fi 6 AX201 160MHz, Python 3.14, StormFiber 5 GHz network. All 7 tutorial steps passed with stable RSSI readings at -48 dBm. See Tutorial #36 for the full walkthrough and test results.

Vital signs from RSSI: The sensing server now supports breathing rate estimation from RSSI variance patterns (requires stationary subject near AP) and motion classification with confidence scoring. RSSI-based vital sign detection has lower fidelity than ESP32 CSI — it is best for presence detection and coarse motion classification.

macOS WiFi (RSSI Only)

Uses CoreWLAN via a Swift helper binary. macOS Sonoma 14.4+ redacts real BSSIDs; the adapter generates deterministic synthetic MACs so the multi-BSSID pipeline still works.

# Compile the Swift helper (once)
swiftc -O archive/v1/src/sensing/mac_wifi.swift -o mac_wifi

# Run natively
./target/release/sensing-server --source macos --http-port 3000 --ws-port 3001 --tick-ms 500

See ADR-025 for details.

Linux WiFi (RSSI Only)

Uses iw dev <iface> scan to capture RSSI. Requires CAP_NET_ADMIN (root) for active scans; use scan dump for cached results without root.

# Run natively (requires root for active scanning)
sudo ./target/release/sensing-server --source linux --http-port 3000 --ws-port 3001 --tick-ms 500

ESP32-S3 (Full CSI)

Real Channel State Information at 20 Hz with 56-192 subcarriers. Required for pose estimation, vital signs, and through-wall sensing.

# From source
./target/release/sensing-server --source esp32 --udp-port 5005 --http-port 3000 --ws-port 3001

# Docker (use CSI_SOURCE environment variable)
docker run -p 3000:3000 -p 3001:3001 -p 5005:5005/udp -e CSI_SOURCE=esp32 ruvnet/wifi-densepose:latest

The ESP32 nodes stream binary CSI frames over UDP to port 5005. See Hardware Setup for flashing instructions.

ESP32 Multistatic Mesh (Advanced)

For higher accuracy with through-wall tracking, deploy 3-6 ESP32-S3 nodes in a multistatic mesh configuration. Each node acts as both transmitter and receiver, creating multiple sensing paths through the environment.

# Start the aggregator with multistatic mode
./target/release/sensing-server --source esp32 --udp-port 5005 --http-port 3000 --ws-port 3001

The mesh uses a Time-Division Multiplexing (TDM) protocol so nodes take turns transmitting, avoiding self-interference. Key features:

See ADR-029 and ADR-032 for the full design.

Connect Mesh Data to the Dashboard and Observatory

If a standalone aggregator command prints live packets, the ESP32 fleet is already reaching that host. To visualize the same data, stop the standalone aggregator and run sensing-server on that same host and UDP port. The sensing server is the aggregator used by the REST API, WebSocket stream, dashboard, and Observatory.

# From a source build
cd v2
cargo run -p wifi-densepose-sensing-server -- \
  --source esp32 \
  --udp-port 5005 \
  --http-port 3000 \
  --ws-port 3001 \
  --ui-path ../../ui

# Docker
docker run --rm \
  -e CSI_SOURCE=esp32 \
  -p 3000:3000 \
  -p 3001:3001 \
  -p 5005:5005/udp \
  ruvnet/wifi-densepose:latest

Open the UI from the sensing server, not from a local file:

Use these checks before debugging the browser:

curl http://localhost:3000/health
curl http://localhost:3000/api/v1/nodes
curl http://localhost:3000/api/v1/sensing/latest

If the ESP32 nodes are provisioned with --target-ip <AGGREGATOR_HOST>, that IP must be the machine running sensing-server. Only one process can receive UDP :5005 at a time, so leave the standalone hardware aggregator off while the dashboard or Observatory is live.

Cognitum Seed Integration (ADR-069)

Connect an ESP32-S3 to a Cognitum Seed (Pi Zero 2 W, ~$15) for persistent vector storage, kNN similarity search, cryptographic witness chain, and AI-accessible sensing via MCP proxy.

What the Seed adds:

• RVF vector store — Persistent 8-dim feature vectors with content-addressed IDs and kNN search (cosine, L2, dot product)
• Witness chain — SHA-256 tamper-evident audit trail for every ingest operation
• Ed25519 custody — Device-bound keypair for cryptographic attestation of sensing data
• Sensor fusion — BME280 (temp/humidity/pressure), PIR motion, reed switch, 4-ch ADC provide environmental ground truth
• MCP proxy — 114 tools via JSON-RPC 2.0 so AI assistants (Claude, GPT) can query sensing state directly
• Reflex rules — Automatic alarm triggers based on fragility, drift, and anomaly thresholds

Setup:

# 1. Plug in the Cognitum Seed via USB — appears as a network adapter at 169.254.42.1

# 2. Pair your client (opens a 30-second window, USB-only for security)
curl -sk -X POST https://169.254.42.1:8443/api/v1/pair/window
curl -sk -X POST https://169.254.42.1:8443/api/v1/pair \
  -H 'Content-Type: application/json' -d '{"client_name":"my-laptop"}'
# Save the returned token — it is shown only once

# 3. Provision ESP32 to send features to your laptop (where the bridge runs)
python firmware/esp32-csi-node/provision.py --port COM9 \
  --ssid "YourWiFi" --password "secret" \
  --target-ip 192.168.1.20 --target-port 5006 --node-id 1

# 4. Run the bridge (receives ESP32 UDP, ingests into Seed via HTTPS)
export SEED_TOKEN="your-pairing-token"
python scripts/seed_csi_bridge.py \
  --seed-url https://169.254.42.1:8443 --token "$SEED_TOKEN" \
  --udp-port 5006 --batch-size 10 --validate

# 5. Check Seed status
python scripts/seed_csi_bridge.py --token "$SEED_TOKEN" --stats

# 6. Trigger compaction (reclaim disk space from deleted vectors)
python scripts/seed_csi_bridge.py --token "$SEED_TOKEN" --compact

Feature vector dimensions (magic 0xC5110003, 48 bytes, 1 Hz):

Architecture:

ESP32-S3 ($9)  ──UDP:5006──>  Host (bridge)  ──HTTPS──>  Cognitum Seed ($15)
  CSI @ 100 Hz                seed_csi_bridge.py           RVF vector store
  Features @ 1 Hz            Batches, validates            kNN graph + boundary
  Vitals @ 1 Hz              NaN rejection                 Witness chain
                              Source IP filtering           114-tool MCP proxy

See ADR-069 for the complete design, validation results, and security analysis.

REST API Reference

Base URL: http://localhost:3000 (Docker) or http://localhost:8080 (binary default).

Example: Get fleet mesh state (ADR-110)

curl -s http://localhost:3000/api/v1/mesh | python -m json.tool

{
    "nodes": {
        "9": {
            "offset_us":       1163565,
            "is_leader":       false,
            "is_valid":        true,
            "smoothed":        true,
            "sequence":        20,
            "csi_fps_ema":     10.0,
            "csi_fps_samples": 47
        },
        "12": {
            "offset_us":       -7,
            "is_leader":       true,
            "is_valid":        true,
            "smoothed":        false,
            "sequence":        20,
            "csi_fps_ema":     10.0,
            "csi_fps_samples": 51
        }
    },
    "total": 2
}

Empty {"nodes": {}, "total": 0} means no mesh peers reachable. Nodes that haven't emitted a sync packet yet are omitted from the map.

Example: Get one node's sync state

curl -s http://localhost:3000/api/v1/nodes/9/sync | python -m json.tool

200 → same NodeSyncSnapshot shape as inside /api/v1/mesh or the WebSocket sync field. Field meanings are documented under Per-node mesh sync (ADR-110).

404 (unknown node):

{"error": "unknown_node", "node_id": 99}

404 (node exists but hasn't synced yet):

{
    "error":   "no_sync",
    "node_id": 9,
    "hint":    "node hasn't emitted a sync packet yet (no mesh peer or not v0.6.9+)"
}

Useful for Home Assistant REST sensors, Prometheus exporters, automation rule probes, and curl debugging — anywhere you want one-shot mesh state without holding a WebSocket connection.

Example: Get Vital Signs

curl -s http://localhost:3000/api/v1/vital-signs | python -m json.tool

{
    "breathing_bpm": 16.2,
    "heart_bpm": 72.1,
    "breathing_confidence": 0.87,
    "heart_confidence": 0.63,
    "motion_level": 0.12,
    "timestamp_ms": 1709312400000
}

Example: Get Pose

curl -s http://localhost:3000/api/v1/pose/current | python -m json.tool

{
    "persons": [
        {
            "id": 0,
            "keypoints": [
                {"name": "nose", "x": 0.52, "y": 0.31, "z": 0.0, "confidence": 0.91},
                {"name": "left_eye", "x": 0.54, "y": 0.29, "z": 0.0, "confidence": 0.88}
            ]
        }
    ],
    "frame_id": 1024,
    "timestamp_ms": 1709312400000
}

WebSocket Streaming

Real-time sensing data is available via WebSocket.

URL: ws://localhost:3000/ws/sensing (same port as HTTP — recommended) or ws://localhost:3001/ws/sensing (dedicated WS port).

Note: The /ws/sensing WebSocket endpoint is available on both the HTTP port (3000) and the dedicated WebSocket port (3001/8765). The web UI uses the HTTP port so only one port needs to be exposed. The dedicated WS port remains available for backward compatibility.

Python Example

import asyncio
import websockets
import json

async def stream():
    uri = "ws://localhost:3001/ws/sensing"
    async with websockets.connect(uri) as ws:
        async for message in ws:
            data = json.loads(message)
            persons = data.get("persons", [])
            vitals = data.get("vital_signs", {})
            print(f"Persons: {len(persons)}, "
                  f"Breathing: {vitals.get('breathing_bpm', 'N/A')} BPM")

asyncio.run(stream())

JavaScript Example

const ws = new WebSocket("ws://localhost:3001/ws/sensing");

ws.onmessage = (event) => {
    const data = JSON.parse(event.data);
    console.log("Persons:", data.persons?.length ?? 0);
    console.log("Breathing:", data.vital_signs?.breathing_bpm, "BPM");
};

ws.onerror = (err) => console.error("WebSocket error:", err);

curl (single frame)

# Requires wscat (npm install -g wscat)
wscat -c ws://localhost:3001/ws/sensing

Per-node mesh sync (ADR-110)

Since firmware v0.7.0-esp32 + sensing-server iter 23, every sensing_update whose nodes participate in the ADR-110 ESP-NOW mesh carries an optional sync object per node:

{
  "type": "sensing_update",
  "nodes": [
    {
      "node_id": 9,
      "rssi_dbm": -38.0,
      "amplitude": [...],
      "subcarrier_count": 64,
      "sync": {
        "offset_us":       1163565,
        "is_leader":       false,
        "is_valid":        true,
        "smoothed":        true,
        "sequence":        20,
        "csi_fps_ema":     10.0,
        "csi_fps_samples": 47
      }
    }
  ]
}

Field meanings:

When sync is omitted entirely: the node isn't on the mesh (or hasn't heard a peer yet). Non-ESP32 paths — multi-BSSID router scan, synthetic-RSSI fallback, simulation — also omit sync. Existing pre-iter-23 UI clients ignore the new field naturally because they don't read it.

How to render this in a UI:

• is_leader === true → badge the node "Leader"
• is_valid === false → grey out / "Sync lost"
• csi_fps_samples < 5 → label as "Calibrating" until ≥5 frames
• |offset_us| trend → render a jitter histogram to show the §A0.10 EMA suppression working live

How to recover a mesh-aligned timestamp for any CSI frame from this node: take the frame's own sequence u32, subtract sync.sequence, divide by sync.csi_fps_ema (or 20.0 if csi_fps_samples < 5), multiply by 1 000 000 µs — that's the mesh delta from the sync emit time. Use it to align multistatic frames from sibling boards.

Home Assistant + Matter integration

Full design + operator guide: docs/integrations/home-assistant.md (ADR-115).

30-second Mosquitto-add-on flow

1. Inside Home Assistant, install the Mosquitto broker add-on from the Add-on Store and start it.

2. In HA, Settings → Devices & Services → Add Integration → MQTT, point at the broker.

3. Start the sensing-server with MQTT:

   bashCopy

   docker run --rm --net=host ruvnet/wifi-densepose:0.7.0 \
       --source esp32 --mqtt --mqtt-host <ha-host-ip>
   

4. Within ~5 seconds HA auto-creates one device per RuView node with 21 entities: 11 raw signals (presence, person count, HR, BR, motion, fall, RSSI, zones, pose, …) plus 10 semantic primitives (someone-sleeping, possible-distress, room-active, elderly-inactivity-anomaly, meeting, bathroom, fall-risk, bed-exit, no-movement, multi-room-transition).

Privacy mode for healthcare / AAL

sensing-server --mqtt --mqtt-host <broker> --mqtt-tls --privacy-mode

--privacy-mode strips heart rate, breathing rate, and pose keypoints from MQTT and Matter — they never reach the wire. Semantic primitives stay published because they're inferred states server-side, not biometric values. This is the architectural win that makes ADR-115 healthcare- and enterprise-deployable.

Matter Bridge (Apple Home / Google Home / Alexa / SmartThings)

sensing-server --matter --matter-setup-file /var/run/ruview-matter.txt

Open /var/run/ruview-matter.txt for the Matter pairing QR / 11-digit setup code. Scan it from Apple Home / Google Home / your HA Matter integration. RuView appears as a Bridged Device with one occupancy endpoint per node + per zone, plus a momentary switch for fall events.

Detailed entity reference, blueprint catalog, troubleshooting recipe matrix: see docs/integrations/home-assistant.md.

Web UI

The built-in Three.js UI is served at http://localhost:3000/ui/ (Docker) or the configured HTTP port.

Two visualization modes:

Dashboard panels:

Both UIs update in real-time via WebSocket and auto-detect the sensing server on the same origin.

Dense Point Cloud (Camera + WiFi CSI Fusion)

RuView can generate real-time 3D point clouds by fusing camera depth estimation with WiFi CSI spatial sensing. This creates a spatial model of the environment that updates in real-time.

Setup

# Build the pointcloud binary
cd v2
cargo build --release -p wifi-densepose-pointcloud

# Start the server (auto-detects camera + CSI). Loopback-only by default.
./target/release/ruview-pointcloud serve --bind 127.0.0.1:9880

Open http://localhost:9880 for the interactive Three.js 3D viewer.

Security note. The server exposes live camera, skeleton, vitals, and occupancy over HTTP. The --bind flag defaults to 127.0.0.1:9880 (loopback-only). Exposing on 0.0.0.0 or a LAN IP is opt-in — the server logs a warning when it does, but there is no auth/TLS layer. Put a reverse proxy in front if you need remote access.

Brain URL. Observations are POSTed to http://127.0.0.1:9876 by default. Override via the RUVIEW_BRAIN_URL environment variable or the --brain <url> flag on serve / train.

Sensors

Commands

Pipeline Components

1. ADR-018 Parser — Decodes ESP32 CSI binary frames from UDP (magic 0xC5110001 raw CSI and 0xC5110006 feature state), extracts I/Q subcarrier amplitudes and phases. Lives in parser.rs; unit-tested against hand-rolled test vectors.
2. Pose (stub) — 17 COCO keypoint layout generated by heuristic_pose_from_amplitude from CSI amplitude energy. This is not the trained WiFlow model — it is a placeholder so the viewer has a skeleton to render. Wiring to real Candle/ONNX inference from the wifi-densepose-nn crate is a planned follow-up.
3. Vital Signs — Breathing rate from CSI phase analysis (peak counting on stable subcarrier)
4. Motion Detection — CSI amplitude variance over 20 frames, triggers adaptive capture
5. RF Tomography — Backprojection from per-node RSSI to 8×8×4 occupancy grid
6. Camera Depth — MiDaS monocular depth (GPU) with luminance+edge fallback
7. Sensor Fusion — Voxel-grid merging of camera depth + CSI occupancy
8. Brain Bridge — Stores spatial observations in the ruOS brain every 60 seconds

API Endpoints

Training

The training pipeline calibrates depth estimation and occupancy detection:

ruview-pointcloud train --data-dir ~/.local/share/ruview/training --brain http://127.0.0.1:9876

This captures frames, runs depth calibration (grid search over scale/offset/gamma), trains occupancy thresholds, exports DPO preference pairs, and submits results to the ruOS brain.

Output Formats

• PLY — Standard 3D point cloud (ASCII, with RGB color)
• Gaussian Splats — JSON format for Three.js rendering
• Brain Memories — Spatial observations stored as spatial-observation, spatial-motion, spatial-vitals

Deep Room Scan

Capture a high-quality 3D model of the room:

# Stop the live server first (frees the camera)
# Then capture 20 frames and process with MiDaS
ruview-pointcloud capture --frames 20 --output room_model.ply

Result: 40,000+ voxels at 5cm resolution, 12,000+ Gaussian splats.

ESP32 Provisioning for CSI

To send CSI data to the pointcloud server:

python3 firmware/esp32-csi-node/provision.py \
    --port /dev/ttyACM0 \
    --ssid "YourWiFi" --password "YourPassword" \
    --target-ip 192.168.1.123 --target-port 3333 \
    --node-id 1

Vital Sign Detection

The system extracts breathing rate and heart rate from CSI signal fluctuations using FFT peak detection.

Requirements:

• CSI-capable hardware (ESP32-S3 or research NIC) for accurate readings
• Subject within ~3-5 meters of an access point (up to ~8 m with multistatic mesh)
• Relatively stationary subject (large movements mask vital sign oscillations)

Signal smoothing: Vital sign estimates pass through a three-stage smoothing pipeline (ADR-048): outlier rejection (±8 BPM HR, ±2 BPM BR per frame), 21-frame trimmed mean, and EMA with α=0.02. This produces stable readings that hold steady for 5-10+ seconds instead of jumping every frame. See Adaptive Classifier for details.

Simulated mode produces synthetic vital sign data for testing.

CLI Reference

The Rust sensing server binary accepts the following flags:

Common Invocations

# Simulated mode with UI (development)
./target/release/sensing-server --source simulate --http-port 3000 --ws-port 3001 --ui-path ../../ui

# ESP32 hardware mode
./target/release/sensing-server --source esp32 --udp-port 5005

# Windows WiFi RSSI
./target/release/sensing-server --source wifi --tick-ms 500

# Run benchmark
./target/release/sensing-server --benchmark

# Train and export model
./target/release/sensing-server --train --dataset data/ --epochs 100 --save-rvf model.rvf

# Load trained model with progressive loading
./target/release/sensing-server --model model.rvf --progressive

Observatory Visualization

The Observatory is an immersive Three.js visualization that renders WiFi sensing data as a cinematic 3D experience. It features room-scale props, wireframe human figures, WiFi signal animations, and a live data HUD.

URL: http://localhost:3000/ui/observatory.html

Features:

Keyboard shortcuts:

Live data auto-detect: When served by the sensing server, the Observatory probes /health on the same origin and automatically connects via WebSocket. The HUD badge switches from DEMO to LIVE. No configuration needed.

Loading the Pretrained Model from Hugging Face

A pretrained CSI encoder + presence-detection head is published on Hugging Face at ruvnet/wifi-densepose-pretrained. It was trained on 60,630 frames / 610,615 contrastive triplets (12.2M steps, final loss 0.065) and reports 100% presence accuracy and ~164k embeddings/sec on an Apple M4 Pro.

What it ships (and what it does not):

Download

pip install huggingface_hub
huggingface-cli download ruvnet/wifi-densepose-pretrained \
    --local-dir models/wifi-densepose-pretrained

The download yields a small set of files (the .rvf.jsonl is the canonical container the sensing server reads):

models/wifi-densepose-pretrained/
  model.rvf.jsonl       # RVF container (encoder + presence head + lora)
  model.safetensors     # 48 KB — same encoder weights, safetensors format
  model-q4.bin          # 8 KB — recommended quantization for edge
  presence-head.json    # presence classifier head
  config.json           # sona-lora rank=8 alpha=16, target encoder + task_heads

Using the weights

The HF artifact is in JSONL RVF format (one JSON object per line: metadata, encoder, lora). What you can do with it today:

Known gap (tracked): v2/crates/wifi-densepose-sensing-server/src/rvf_container.rs only parses the binary RVF segment format (magic 0x52564653). Pointing --model at model.rvf.jsonl causes the progressive loader to error with invalid magic at offset 0: expected 0x52564653, got 0x7974227B (0x7974227B is the ASCII bytes {"ty… from the JSONL header), and the live pipeline degrades to null output rather than falling back to heuristic mode. Until a JSONL adapter lands (or the model is re-published as binary RVF), run the sensing-server without --model and consume the HF weights from Python or the training pipeline.

# Works today — Python side (training, evaluation, embedding extraction):
python -c "
from safetensors.torch import load_file
state = load_file('models/wifi-densepose-pretrained/model.safetensors')
print({k: tuple(v.shape) for k, v in state.items()})
"

# Sensing server — run heuristic for now:
cargo run -p wifi-densepose-sensing-server --release -- \
    --source esp32 --udp-port 5005 --http-port 3000

See RVF Model Containers for the binary format the loader expects, and Training a Model for using the encoder as a starting point for environment-specific fine-tuning.

Adaptive Classifier

The adaptive classifier (ADR-048) learns your environment's specific WiFi signal patterns from labeled recordings. It replaces static threshold-based classification with a trained logistic regression model that uses 15 features (7 server-computed + 8 subcarrier-derived statistics).

Signal Smoothing Pipeline

All CSI-derived metrics pass through a three-stage pipeline before reaching the UI:

Vital signs use additional stabilization:

Recording Training Data

Record labeled CSI sessions while performing distinct activities. Each recording captures full sensing frames (features + raw subcarrier amplitudes) at ~10-25 FPS.

# 1. Record empty room (leave the room for 30 seconds)
curl -X POST http://localhost:3000/api/v1/recording/start \
  -H "Content-Type: application/json" -d '{"id":"train_empty_room"}'
# ... wait 30 seconds ...
curl -X POST http://localhost:3000/api/v1/recording/stop

# 2. Record sitting still (sit near ESP32 for 30 seconds)
curl -X POST http://localhost:3000/api/v1/recording/start \
  -H "Content-Type: application/json" -d '{"id":"train_sitting_still"}'
# ... wait 30 seconds ...
curl -X POST http://localhost:3000/api/v1/recording/stop

# 3. Record walking (walk around the room for 30 seconds)
curl -X POST http://localhost:3000/api/v1/recording/start \
  -H "Content-Type: application/json" -d '{"id":"train_walking"}'
# ... wait 30 seconds ...
curl -X POST http://localhost:3000/api/v1/recording/stop

# 4. Record active movement (jumping jacks, arm waving for 30 seconds)
curl -X POST http://localhost:3000/api/v1/recording/start \
  -H "Content-Type: application/json" -d '{"id":"train_active"}'
# ... wait 30 seconds ...
curl -X POST http://localhost:3000/api/v1/recording/stop

Recordings are saved as JSONL files in data/recordings/. Filenames must start with train_ and contain a class keyword:

Training the Model

Train the adaptive classifier from your labeled recordings:

curl -X POST http://localhost:3000/api/v1/adaptive/train

The server trains a multiclass logistic regression on 15 features using mini-batch SGD (200 epochs). Training completes in under 1 second for typical recording sets. The trained model is saved to data/adaptive_model.json and automatically loaded on server restart.

Check model status:

curl http://localhost:3000/api/v1/adaptive/status

Unload the model (revert to threshold-based classification):

curl -X POST http://localhost:3000/api/v1/adaptive/unload

Using the Trained Model

Once trained, the adaptive model runs automatically:

1. Each CSI frame is classified using the learned weights instead of static thresholds
2. Model confidence is blended with smoothed threshold confidence (70/30 split)
3. The model persists across server restarts (loaded from data/adaptive_model.json)

Tips for better accuracy:

• Record with clearly distinct activities (actually leave the room for "empty")
• Record 30-60 seconds per activity (more data = better model)
• Re-record and retrain if you move the ESP32 or rearrange the room
• The model is environment-specific — retrain when the physical setup changes

Adaptive Classifier API

Training a Model

The training pipeline is implemented in pure Rust (7,832 lines, zero external ML dependencies).

Step 1: Obtain a Dataset

The system supports two public WiFi CSI datasets:

Download the dataset files and place them in a data/ directory.

Step 2: Train

# From source
./target/release/sensing-server --train --dataset data/ --dataset-type mmfi --epochs 100 --save-rvf model.rvf

# Via Docker (mount your data directory)
# Note: Training mode requires overriding the default entrypoint
docker run --rm \
  -v $(pwd)/data:/data \
  -v $(pwd)/output:/output \
  --entrypoint /app/sensing-server \
  ruvnet/wifi-densepose:latest \
  --train --dataset /data --epochs 100 --export-rvf /output/model.rvf

The pipeline runs 10 phases:

1. Dataset loading (MM-Fi .npy or Wi-Pose .mat)
2. Hardware normalization (Intel 5300 / Atheros / ESP32 -> canonical 56 subcarriers)
3. Subcarrier resampling (114->56 or 30->56 via Catmull-Rom interpolation)
4. Graph transformer construction (17 COCO keypoints, 16 bone edges)
5. Cross-attention training (CSI features -> body pose)
6. Domain-adversarial training (MERIDIAN: gradient reversal + virtual domain augmentation)
7. Composite loss optimization (MSE + CE + UV + temporal + bone + symmetry)
8. SONA adaptation (micro-LoRA + EWC++)
9. Sparse inference optimization (hot/cold neuron partitioning)
10. RVF model packaging

Step 3: Use the Trained Model

./target/release/sensing-server --model model.rvf --progressive --source esp32

Progressive loading enables instant startup (Layer A loads in <5ms with basic inference), with full model loading in the background.

Cross-Environment Adaptation (MERIDIAN)

Models trained in one room typically lose 40-70% accuracy in a new room due to different WiFi multipath patterns. The MERIDIAN system (ADR-027) solves this with a 10-second automatic calibration:

1. Deploy the trained model in a new room
2. Collect ~200 unlabeled CSI frames (10 seconds at 20 Hz)
3. The system automatically generates environment-specific LoRA weights via contrastive test-time training
4. No labels, no retraining, no user intervention

MERIDIAN components (all pure Rust, +12K parameters):

See ADR-027 for the full design.

CRV Signal-Line Protocol

The CRV (Coordinate Remote Viewing) signal-line protocol (ADR-033) maps a 6-stage cognitive sensing methodology onto WiFi CSI processing. This enables structured anomaly classification and multi-person disambiguation.

# Enable CRV in your Cargo.toml
cargo add wifi-densepose-ruvector --features crv

See ADR-033 for the full design.

RVF Model Containers

The RuVector Format (RVF) packages a trained model into a single self-contained binary file.

Export

./target/release/sensing-server --export-rvf model.rvf

Load

./target/release/sensing-server --model model.rvf --progressive

Contents

An RVF file contains: model weights, HNSW vector index, quantization codebooks, SONA adaptation profiles, Ed25519 training proof, and vital sign filter parameters.

Deployment Targets

Hardware Setup

Supported targets

The same firmware/esp32-csi-node source tree builds for both. ESP-IDF picks up sdkconfig.defaults.esp32c6 automatically when the target is set to esp32c6; otherwise it uses sdkconfig.defaults (S3). All C6-only modules are #ifdef-gated, so the S3 build is byte-identical to today.

ESP32-S3 Mesh

A 3-6 node ESP32-S3 mesh provides full CSI at 20 Hz. Total cost: ~$54 for a 3-node setup.

What you need:

• 3-6x ESP32-S3 development boards (~$8 each)
• A WiFi router (the CSI source)
• A computer running the sensing server (aggregator)

Flashing firmware:

Pre-built binaries are available at Releases:

Important: Always use v0.4.3.1 or later. Earlier versions have false fall detection alerts (v0.4.2 and below) and CSI disabled in the build config (pre-v0.4.1).

# Flash an ESP32-S3 with 8MB flash (most boards)
python -m esptool --chip esp32s3 --port COM7 --baud 460800 \
  write-flash --flash-mode dio --flash-size 8MB --flash-freq 80m \
  0x0 bootloader.bin 0x8000 partition-table.bin \
  0xf000 ota_data_initial.bin 0x20000 esp32-csi-node.bin

4MB flash boards (e.g. ESP32-S3 SuperMini 4MB): download esp32-csi-node-s3-4mb.bin + partition-table-s3-4mb.bin from the v0.6.7 release (882 KB binary, 52 % partition slack) and use --flash-size 4MB:

python -m esptool --chip esp32s3 --port COM7 --baud 460800 \
  write-flash --flash-mode dio --flash-size 4MB --flash-freq 80m \
  0x0 bootloader.bin 0x8000 partition-table-4mb.bin \
  0xF000 ota_data_initial.bin 0x20000 esp32-csi-node-4mb.bin

Provisioning:

python firmware/esp32-csi-node/provision.py --port COM7 \
  --ssid "YourWiFi" --password "YourPassword" --target-ip 192.168.1.20

Replace 192.168.1.20 with the IP of the machine running the sensing server.

Mesh key provisioning (secure mode):

For multistatic mesh deployments with authenticated beacons (ADR-032), provision a shared mesh key:

python firmware/esp32-csi-node/provision.py --port COM7 \
  --ssid "YourWiFi" --password "YourPassword" --target-ip 192.168.1.20 \
  --mesh-key "$(openssl rand -hex 32)"

All nodes in a mesh must share the same 256-bit mesh key for HMAC-SHA256 beacon authentication. The key is stored in ESP32 NVS flash and zeroed on firmware erase.

ESP32-C6 (Wi-Fi 6 + 802.15.4 research target — ADR-110)

The C6 build adds four capabilities to the existing csi-node firmware, all opt-in via idf.py menuconfig → ESP32-C6 capabilities (ADR-110):

Build + flash:

cd firmware/esp32-csi-node
idf.py set-target esp32c6
idf.py build                    # ~1.0 MB binary, 46% partition slack on 4 MB flash
idf.py -p COM6 flash
# Then provision the same way as S3 (provision.py works for both targets):
python provision.py --port COM6 --ssid "YourWiFi" --password "secret" --target-ip 192.168.1.20

Verifying the C6 modules came up — idf.py -p COM6 monitor should show:

I (353) main: ESP32-C6 CSI Node (ADR-018 / ADR-110) — v0.6.7 — Node ID: 1
I (413) c6_ts: init done: channel=15 EUI=<your-EUI64> leader=yes(candidate)
I (463) wifi: mac_version:HAL_MAC_ESP32AX_761      ← 802.11ax MAC firmware loaded

The c6_ts: init done line confirms the 802.15.4 stack is up; if TWT succeeds you'll also see an iTWT setup event received from AP line after the WiFi connect completes.

Multi-room time-aligned multistatic capture (preview):

Flash two or more C6 boards, leave them on the same 802.15.4 channel (default 15). One will elect itself leader (lowest EUI-64) and broadcast TS_BEACON frames every 100 ms; the others compute and apply offsets. Each CSI frame from a follower carries a c6_timesync_get_epoch_us() wall-clock estimate aligned to within ±100 µs of the leader's monotonic time. Target use case: ADR-029/030 multistatic fusion without burning WiFi airtime on coordination.

Battery seed-node mode (v0.6.7 — real LP-core program):

# Enable LP-core hibernation in menuconfig:
#   ESP32-C6 capabilities (ADR-110) → Enable LP-core wake-on-motion hibernation
#   → LP-core wake GPIO (default 4 �� connect a PIR or accelerometer INT line here)
#   → LP-core poll period (default 10 ms)
#   → LP-core debounce sample count (default 3 consecutive matches)
idf.py menuconfig
idf.py build flash

When enabled, the C6 LP RISC-V coprocessor runs a real polling program (firmware/esp32-csi-node/main/lp_core/main.c) that polls the wake GPIO at the configured cadence, debounces N consecutive matching reads, and wakes the HP core via ulp_lp_core_wakeup_main_processor(). esp_sleep_get_wakeup_cause() returns ESP_SLEEP_WAKEUP_ULP, and c6_lp_core_motion_count() / c6_lp_core_poll_count() expose the LP-side counters for the witness harness. Target standby current ~5 µA (datasheet; pending INA measurement).

Two-board iTWT bench (v0.6.7 — soft-AP HE/TWT, no router required):

Pair two C6 boards — one acts as the iTWT-capable AP, the other as the STA that negotiates and benchmarks the TWT agreement.

# Board #1 (AP role): append to sdkconfig.defaults.esp32c6:
CONFIG_C6_SOFTAP_HE_ENABLE=y
CONFIG_C6_SOFTAP_HE_SSID="ruview-c6-twt"
CONFIG_C6_SOFTAP_HE_PSK="ruviewtwt"
CONFIG_C6_SOFTAP_HE_CHANNEL=6

idf.py set-target esp32c6 && idf.py build && idf.py -p COM6 flash

Board #1 boots in WIFI_MODE_APSTA, advertising HE capabilities and TWT Responder=1 on channel 6. Board #2 provisions to associate with that SSID:

python firmware/esp32-csi-node/provision.py --port COM9 \
  --ssid "ruview-c6-twt" --password "ruviewtwt" --target-ip 192.168.1.20

Board #2 runs the existing c6_twt_setup_default() on connect and now negotiates a real iTWT agreement against the cooperative AP — the iTWT setup queued: wake_interval=10000 µs log line should be followed by an iTWT setup event received from AP instead of the INVALID_ARG graceful fallback that fired against the bench's 11n-only ruv.net AP.

NVS overrides for AP role (namespace ruview): softap_ssid, softap_psk, softap_chan — provision once and the values survive firmware updates.

What's NOT on the C6 build (vs S3 production): no AMOLED display (ADR-045 needs 8 MB + LCD touch driver), no WASM3 (ADR-040 needs PSRAM), no Seeed mmWave fusion (separate board). The C6 is a research/seed target, not a drop-in replacement for the S3 production node.

TDM slot assignment:

Each node in a multistatic mesh needs a unique TDM slot ID (0-based):

# Node 0 (slot 0) — first transmitter
python firmware/esp32-csi-node/provision.py --port COM7 --tdm-slot 0 --tdm-total 3

# Node 1 (slot 1)
python firmware/esp32-csi-node/provision.py --port COM8 --tdm-slot 1 --tdm-total 3

# Node 2 (slot 2)
python firmware/esp32-csi-node/provision.py --port COM9 --tdm-slot 2 --tdm-total 3

Edge Intelligence (v0.3.0-alpha, ADR-039):

The v0.3.0-alpha firmware adds on-device signal processing that runs directly on the ESP32-S3 — no host PC needed for basic presence and vital signs. Edge processing is disabled by default for full backward compatibility.

Enable via NVS (no reflash needed):

# Enable Tier 2 (full vitals) on an already-flashed node
python firmware/esp32-csi-node/provision.py --port COM7 \
  --ssid "YourWiFi" --password "YourPassword" --target-ip 192.168.1.20 \
  --edge-tier 2

Key NVS settings for edge processing:

When Tier 2 is active, the node sends a 32-byte vitals packet at 1 Hz (configurable) containing presence state, motion score, breathing BPM, heart rate BPM, confidence values, fall flag, and occupancy estimate. The packet uses magic 0xC5110002 and is sent to the same aggregator IP and port as raw CSI frames.

Binary size: 990 KB (8MB flash, 52% free) or 773 KB (4MB flash). v0.5.0 adds mmWave sensor fusion (~12 KB larger).

Alpha notice: Vital sign estimation uses heuristic BPM extraction. Accuracy is best with stationary subjects in controlled environments. Not for medical use.

Start the aggregator:

# From source
./target/release/sensing-server --source esp32 --udp-port 5005 --http-port 3000 --ws-port 3001

# Docker (use CSI_SOURCE environment variable)
docker run -p 3000:3000 -p 3001:3001 -p 5005:5005/udp -e CSI_SOURCE=esp32 ruvnet/wifi-densepose:latest

See ADR-018, ADR-029, and Tutorial #34.

Intel 5300 / Atheros NIC

These research NICs provide full CSI on Linux with firmware/driver modifications.

These are advanced setups. See the respective driver documentation for installation.

Camera-Free Pose Training

RuView can train a 17-keypoint COCO pose model without any camera by fusing 10 sensor signals from the ESP32 nodes and Cognitum Seed:

How It Works

The pipeline generates weak labels from sensor fusion, then trains in 5 phases:

1. Multi-modal collection — Syncs CSI frames with Seed sensor events
2. Weak label generation — RSSI triangulation for head position, subcarrier asymmetry for hands, vibration for feet
3. 5-keypoint pose proxy — Trains head/hands/feet positions from fused signals
4. 17-keypoint interpolation — Derives full COCO skeleton using bone length constraints
5. Self-refinement — Bootstraps from confident predictions (3 rounds)

# With Cognitum Seed connected (all 10 signals):
node scripts/train-camera-free.js \
  --data data/recordings/pretrain-*.csi.jsonl \
  --seed-url https://169.254.42.1:8443 \
  --seed-token "$SEED_TOKEN"

# Without Seed (CSI-only, 3 signals — still works):
node scripts/train-camera-free.js \
  --data data/recordings/pretrain-*.csi.jsonl --no-seed

Output: 82.8 KB model (8 KB at 4-bit) with 17-keypoint predictions, 0 skeleton violations, LoRA per-node adapters, and EWC protection against forgetting.

See ADR-071 and the pretraining tutorial for the full walkthrough.

Camera-Supervised Pose Training (v0.7.0)

For significantly higher accuracy, use a webcam as a temporary teacher during training. The camera captures real 17-keypoint poses via MediaPipe, paired with simultaneous ESP32 CSI data. After training, the camera is no longer needed — the model runs on CSI only.

Result: 92.9% PCK@20 from a 5-minute collection session.

Requirements

• Python 3.9+ with mediapipe and opencv-python (pip install mediapipe opencv-python)
• ESP32-S3 node streaming CSI over UDP (port 5005)
• A webcam (laptop, USB, or Mac camera via Tailscale)

Step 1: Capture Camera + CSI Simultaneously

Run both scripts at the same time (in separate terminals):

# Terminal 1: Record ESP32 CSI
python scripts/record-csi-udp.py --duration 300

# Terminal 2: Capture camera keypoints
python scripts/collect-ground-truth.py --duration 300 --preview

Move around naturally in front of the camera for 5 minutes. The --preview flag shows a live skeleton overlay.

Step 2: Align and Train

# Align camera keypoints with CSI windows
node scripts/align-ground-truth.js \
  --gt data/ground-truth/*.jsonl \
  --csi data/recordings/csi-*.csi.jsonl

# Train (start with lite, scale up as you collect more data)
node scripts/train-wiflow-supervised.js \
  --data data/paired/*.jsonl \
  --scale lite \
  --epochs 50

# Evaluate
node scripts/eval-wiflow.js \
  --model models/wiflow-supervised/wiflow-v1.json \
  --data data/paired/*.jsonl

Scale Presets

See ADR-079 for the full design and optimization details.

Pre-Trained Models (No Training Required)

Pre-trained models are available on HuggingFace: https://huggingface.co/ruvnet/wifi-densepose-pretrained

Download and start sensing immediately — no datasets, no GPU, no training needed.

Quick Start with Pre-Trained Models

# Install huggingface CLI
pip install huggingface_hub

# Download all models
huggingface-cli download ruvnet/wifi-densepose-pretrained --local-dir models/pretrained

# The models include:
#   model.safetensors    — 48 KB contrastive encoder
#   model-q4.bin         — 8 KB quantized (recommended)
#   model-q2.bin         — 4 KB ultra-compact (ESP32 edge)
#   presence-head.json   — presence detection head (100% accuracy)
#   node-1.json          — LoRA adapter for room 1
#   node-2.json          — LoRA adapter for room 2

What the Models Do

The pre-trained encoder converts 8-dim CSI feature vectors into 128-dim embeddings. These embeddings power all 17 sensing applications:

• Presence detection — 100% accuracy, never misses, never false alarms
• Environment fingerprinting — kNN search finds "states like this one"
• Anomaly detection — embeddings that don't match known clusters = anomaly
• Activity classification — different activities cluster in embedding space
• Room adaptation — swap LoRA adapters for different rooms without retraining

Retraining on Your Own Data

If you want to improve accuracy for your specific environment:

# Collect 2+ minutes of CSI from your ESP32
python scripts/collect-training-data.py --port 5006 --duration 120

# Retrain (uses ruvllm, no PyTorch needed)
node scripts/train-ruvllm.js --data data/recordings/*.csi.jsonl

# Benchmark your retrained model
node scripts/benchmark-ruvllm.js --model models/csi-ruvllm

Health & Wellness Applications

WiFi sensing can monitor health metrics without any wearable or camera:

# Sleep quality monitoring (run overnight)
node scripts/sleep-monitor.js --port 5006 --bind 192.168.1.20

# Breathing disorder pre-screening
node scripts/apnea-detector.js --port 5006 --bind 192.168.1.20

# Stress detection via heart rate variability
node scripts/stress-monitor.js --port 5006 --bind 192.168.1.20

# Walking analysis + tremor detection
node scripts/gait-analyzer.js --port 5006 --bind 192.168.1.20

# Replay on recorded data (no live hardware needed)
node scripts/sleep-monitor.js --replay data/recordings/*.csi.jsonl

Note: These are pre-screening tools, not medical devices. Consult a healthcare professional for diagnosis.

ruvllm Training Pipeline

All training uses ruvllm — a Rust-native ML runtime. No Python, no PyTorch, no GPU drivers required. Runs on any machine with Node.js.

5-Phase Training

# Basic training
node scripts/train-ruvllm.js --data data/recordings/pretrain-*.csi.jsonl

# Benchmark
node scripts/benchmark-ruvllm.js --model models/csi-ruvllm

Quantization Options

Key Features

• SONA adaptation — Adapts to new rooms in <1ms without retraining
• LoRA adapters — 2,048 parameters per room, hot-swappable
• EWC protection — Learns new rooms without forgetting previous ones
• Deterministic — Same seed always produces same model (reproducible)
• 10x data augmentation — Temporal interpolation, noise injection, cross-node blending

Docker Compose (Multi-Service)

For production deployments with both Rust and Python services:

cd docker
docker compose up

This starts:

• Rust sensing server on ports 3000 (HTTP), 3001 (WS), 5005 (UDP)
• Python legacy server on ports 8080 (HTTP), 8765 (WS)

Testing Firmware Without Hardware (QEMU)

You can test the ESP32-S3 firmware on your computer without any physical hardware. The project uses QEMU — an emulator that pretends to be an ESP32-S3 chip, running the real firmware code inside a virtual machine on your PC.

This is useful when:

• You don't have an ESP32-S3 board yet
• You want to test firmware changes before flashing to real hardware
• You're running automated tests in CI/CD
• You want to simulate multiple ESP32 nodes talking to each other

What You Need

Required:

• Python 3.8+ (you probably already have this)
• QEMU with ESP32-S3 support (Espressif's fork)

Install QEMU (one-time setup):

# Easiest: use the automated installer (installs QEMU + Python tools)
bash scripts/install-qemu.sh

# Or check what's already installed:
bash scripts/install-qemu.sh --check

The installer detects your OS (Ubuntu, Fedora, macOS, etc.), installs build dependencies, clones Espressif's QEMU fork, builds it, and adds it to your PATH. It also installs the Python tools (esptool, pyyaml, esp-idf-nvs-partition-gen).

# Build from source
git clone https://github.com/espressif/qemu.git
cd qemu
./configure --target-list=xtensa-softmmu --enable-slirp
make -j$(nproc)
export QEMU_PATH=$(pwd)/build/qemu-system-xtensa

# Install Python tools
pip install esptool pyyaml esp-idf-nvs-partition-gen

For multi-node testing (optional):

# Linux only — needed for virtual network bridges
sudo apt install socat bridge-utils iproute2

The qemu-cli.sh Command

All QEMU testing is available through a single command:

bash scripts/qemu-cli.sh <command>

Your First Test Run

The simplest way to test the firmware:

# Using the CLI:
bash scripts/qemu-cli.sh test

# Or directly:
bash scripts/qemu-esp32s3-test.sh

What happens behind the scenes:

1. The firmware is compiled with a "mock CSI" mode — instead of reading real WiFi signals, it generates synthetic test data that mimics real people walking, falling, or breathing
2. The compiled firmware is loaded into QEMU, which boots it like a real ESP32-S3
3. The emulator's serial output (what you'd see on a USB cable) is captured
4. A validation script checks the output for expected behavior and errors

If you already built the firmware and want to skip rebuilding:

SKIP_BUILD=1 bash scripts/qemu-esp32s3-test.sh

To give it more time (useful on slower machines):

QEMU_TIMEOUT=120 bash scripts/qemu-esp32s3-test.sh

Understanding the Test Output

The test runs 16 checks on the firmware's output. Here's what a successful run looks like:

=== QEMU ESP32-S3 Firmware Test (ADR-061) ===

[PASS] Boot: Firmware booted successfully
[PASS] NVS config: Configuration loaded from flash
[PASS] Mock CSI: Synthetic WiFi data generator started
[PASS] Edge processing: Signal analysis pipeline running
[PASS] Frame serialization: Data packets formatted correctly
[PASS] No crashes: No error conditions detected
...

16/16 checks passed
=== Test Complete (exit code: 0) ===

Exit codes explained:

Testing Multiple Nodes at Once (Swarm)

Real deployments use 3-8 ESP32 nodes. The swarm configurator lets you simulate multiple nodes on your computer, each with a different role:

• Sensor nodes — generate WiFi signal data (like ESP32s placed around a room)
• Coordinator node — collects data from all sensors and runs analysis
• Gateway node — bridges data to your computer

# Quick 2-node smoke test (15 seconds)
python3 scripts/qemu_swarm.py --preset smoke

# Standard 3-node test: 2 sensors + 1 coordinator (60 seconds)
python3 scripts/qemu_swarm.py --preset standard

# See what's available
python3 scripts/qemu_swarm.py --list-presets

# Preview what would run (without actually running)
python3 scripts/qemu_swarm.py --preset standard --dry-run

Note: Multi-node testing with virtual bridges requires Linux and sudo. On other systems, nodes use a simpler networking mode where each node can reach the coordinator but not each other.

Swarm Presets

Writing Your Own Swarm Config

Create a YAML file describing your test scenario:

# my_test.yaml
swarm:
  name: my-custom-test
  duration_s: 45
  topology: star       # star, mesh, line, or ring
  aggregator_port: 5005

nodes:
  - role: coordinator
    node_id: 0
    scenario: 0        # 0=empty room (baseline)
    channel: 6
    edge_tier: 2

  - role: sensor
    node_id: 1
    scenario: 2        # 2=walking person
    channel: 6
    tdm_slot: 1

  - role: sensor
    node_id: 2
    scenario: 3        # 3=fall event
    channel: 6
    tdm_slot: 2

assertions:
  - all_nodes_boot           # Did every node start up?
  - no_crashes               # Any error/panic?
  - all_nodes_produce_frames # Is each sensor generating data?
  - fall_detected_by_node_2  # Did node 2 detect the fall?

Available scenarios (what kind of fake WiFi data to generate):

Topology options:

Run your custom config:

python3 scripts/qemu_swarm.py --config my_test.yaml

Debugging Firmware in QEMU

If something goes wrong, you can attach a debugger to the emulated ESP32:

# Terminal 1: Start QEMU with debug support (paused at boot)
qemu-system-xtensa -machine esp32s3 -nographic \
  -drive file=firmware/esp32-csi-node/build/qemu_flash.bin,if=mtd,format=raw \
  -s -S

# Terminal 2: Connect the debugger
xtensa-esp-elf-gdb firmware/esp32-csi-node/build/esp32-csi-node.elf \
  -ex "target remote :1234" \
  -ex "break app_main" \
  -ex "continue"

Or use VS Code: open the project, press F5, and select "QEMU ESP32-S3 Debug".

Running the Full Test Suite

For thorough validation before submitting a pull request:

# 1. Single-node test (2 minutes)
bash scripts/qemu-esp32s3-test.sh

# 2. Multi-node swarm test (1 minute)
python3 scripts/qemu_swarm.py --preset standard

# 3. Fuzz testing — finds edge-case crashes (1-5 minutes)
cd firmware/esp32-csi-node/test
make all CC=clang
make run_serialize FUZZ_DURATION=60
make run_edge FUZZ_DURATION=60
make run_nvs FUZZ_DURATION=60

# 4. NVS configuration matrix — tests 14 config combinations
python3 scripts/generate_nvs_matrix.py --output-dir build/nvs_matrix

# 5. Chaos testing — injects faults to test resilience (2 minutes)
bash scripts/qemu-chaos-test.sh

All of these also run automatically in CI when you push changes to firmware/.

Troubleshooting

Docker: "no matching manifest for linux/arm64" on macOS

The latest tag supports both amd64 and arm64. Pull the latest image:

docker pull ruvnet/wifi-densepose:latest

If you still see this error, your local Docker may have a stale cached manifest. Try:

docker pull --platform linux/arm64 ruvnet/wifi-densepose:latest

Docker: "Connection refused" on localhost:3000

Make sure you're mapping the ports correctly:

docker run -p 3000:3000 -p 3001:3001 ruvnet/wifi-densepose:latest

The -p 3000:3000 maps host port 3000 to container port 3000.

Docker: No WebSocket data in UI

Add the WebSocket port mapping:

docker run -p 3000:3000 -p 3001:3001 ruvnet/wifi-densepose:latest

ESP32: "CSI not enabled in menuconfig"

Firmware versions prior to v0.4.1 had CONFIG_ESP_WIFI_CSI_ENABLED disabled in the build config. Upgrade to v0.4.1 or later. If building from source, ensure sdkconfig.defaults exists (not just sdkconfig.defaults.template). See ADR-057.

ESP32: No data arriving

1. Verify firmware is v0.4.1+ (older versions had CSI disabled — see above)
2. Verify the ESP32 is connected to the same WiFi network
3. Check the target IP matches the sensing server machine: python firmware/esp32-csi-node/provision.py --port COM7 --target-ip <YOUR_IP>
4. Verify UDP port 5005 is not blocked by firewall
5. Test with: nc -lu 5005 (Linux) or similar UDP listener

Build: Rust compilation errors

Ensure Rust 1.75+ is installed (1.85+ recommended):

rustup update stable
rustc --version

Build: Linux native desktop prerequisites

If you are compiling the Rust workspace on a Debian/Ubuntu-based Linux system, install the native desktop development packages first:

sudo apt update
sudo apt install -y \
  build-essential pkg-config \
  libglib2.0-dev libgtk-3-dev \
  libsoup-3.0-dev \
  libjavascriptcoregtk-4.1-dev \
  libwebkit2gtk-4.1-dev

Then rerun:

cargo build --release

This is the same Linux pre-step referenced in the Rust source build section and covers the common GTK/WebKit pkg-config requirements used by the desktop build.

Windows: RSSI mode shows no data

Run the terminal as Administrator (required for netsh wlan access). Verified working on Windows 10 and 11 with Intel AX201 and Intel BE201 adapters.

Vital signs show 0 BPM

• Vital sign detection requires CSI-capable hardware (ESP32 or research NIC)
• RSSI-only mode (Windows WiFi) does not have sufficient resolution for vital signs
• In simulated mode, synthetic vital signs are generated after a few seconds of warm-up
• With real ESP32 data, vital signs take ~5 seconds to stabilize (smoothing pipeline warm-up)

Vital signs jumping around

The server applies a 3-stage smoothing pipeline (ADR-048). If readings are still unstable:

• Ensure the subject is relatively still (large movements mask vital sign oscillations)
• Train the adaptive classifier for your specific environment: curl -X POST http://localhost:3000/api/v1/adaptive/train
• Check signal quality: curl http://localhost:3000/api/v1/sensing/latest — look for signal_quality > 0.4

Observatory shows DEMO instead of LIVE

• Verify the sensing server is running: curl http://localhost:3000/health
• Access Observatory via the server URL: http://localhost:3000/ui/observatory.html (not a file:// URL)
• If a standalone aggregator command is already listening on UDP :5005, stop it and run sensing-server --source esp32 --udp-port 5005 instead; the Observatory reads the server WebSocket, not the standalone aggregator output
• Verify the ESP32 nodes are provisioned to the IP address of the machine running sensing-server
• Hard refresh with Ctrl+Shift+R to clear cached settings
• The auto-detect probes /health on the same origin — cross-origin won't work

QEMU: "qemu-system-xtensa: command not found"

QEMU for ESP32-S3 must be built from Espressif's fork — it is not in standard package managers:

git clone https://github.com/espressif/qemu.git
cd qemu && ./configure --target-list=xtensa-softmmu && make -j$(nproc)
export QEMU_PATH=$(pwd)/build/qemu-system-xtensa

Or point to an existing build: QEMU_PATH=/path/to/qemu-system-xtensa bash scripts/qemu-esp32s3-test.sh

QEMU: Test times out with no output

The emulator is slower than real hardware. Increase the timeout:

QEMU_TIMEOUT=120 bash scripts/qemu-esp32s3-test.sh

If there's truly no output at all, the firmware build may have failed. Rebuild without SKIP_BUILD:

bash scripts/qemu-esp32s3-test.sh   # without SKIP_BUILD

QEMU: "esptool not found"

Install it with pip: pip install esptool

QEMU Swarm: "Must be run as root"

Multi-node swarm tests with virtual network bridges require root on Linux. Two options:

1. Run with sudo: sudo python3 scripts/qemu_swarm.py --preset standard
2. Skip bridges (nodes use simpler networking): the tool automatically falls back on non-root systems, but nodes can't communicate with each other (only with the aggregator)

QEMU Swarm: "yaml module not found"

Install PyYAML: pip install pyyaml

FAQ

Q: Do I need special hardware to try this? No. Run docker run -p 3000:3000 ruvnet/wifi-densepose:latest and open http://localhost:3000. Simulated mode exercises the full pipeline with synthetic data.

Q: Can consumer WiFi laptops do pose estimation? No. Consumer WiFi exposes only RSSI (one number per access point), not CSI (56+ complex subcarrier values per frame). RSSI supports coarse presence and motion detection. Full pose estimation requires CSI-capable hardware like an ESP32-S3 ($8) or a research NIC.

Q: How accurate is the pose estimation? Accuracy depends on hardware and environment. With a 3-node ESP32 mesh in a single room, the system tracks 17 COCO keypoints. The core algorithm follows the CMU "DensePose From WiFi" paper (arXiv:2301.00250). The MERIDIAN domain generalization system (ADR-027) reduces cross-environment accuracy loss from 40-70% to under 15% via 10-second automatic calibration.

Q: Does it work through walls? Yes. WiFi signals penetrate non-metallic materials (drywall, wood, concrete up to ~30cm). Metal walls/doors significantly attenuate the signal. With a single AP the effective through-wall range is approximately 5 meters. With a 3-6 node multistatic mesh (ADR-029), attention-weighted cross-viewpoint fusion extends the effective range to ~8 meters through standard residential walls.

Q: How many people can it track? Each access point can distinguish ~3-5 people with 56 subcarriers. Multi-AP deployments multiply linearly (e.g., 4 APs cover ~15-20 people). There is no hard software limit; the practical ceiling is signal physics.

Q: Is this privacy-preserving? The system uses WiFi radio signals, not cameras. No images or video are captured or stored. However, it does track human position, movement, and vital signs, which is personal data subject to applicable privacy regulations.

Q: What's the Python vs Rust difference? The Rust implementation (v2) is 810x faster than Python (v1) for the full CSI pipeline. The Docker image is 132 MB vs 569 MB. Rust is the primary and recommended runtime. Python v1 remains available for legacy workflows.

Q: Can I use an ESP8266 instead of ESP32-S3? No. The ESP8266 does not expose WiFi Channel State Information (CSI) through its SDK, has insufficient RAM (~80 KB vs 512 KB), and runs a single-core 80 MHz CPU that cannot handle the signal processing pipeline. The ESP32-S3 is the minimum supported CSI capture device. See Issue #138 for alternatives including using cheap Android TV boxes as aggregation hubs.

Q: Does the Windows WiFi tutorial work on Windows 10? Yes. Community-tested on Windows 10 (build 26200) with an Intel Wi-Fi 6 AX201 160MHz adapter on a 5 GHz network. All 7 tutorial steps passed with Python 3.14. See Issue #36 for full test results.

Q: Can I run the sensing server on an ARM device (Raspberry Pi, TV box)? ARM64 deployment is planned (ADR-046) but not yet available as a pre-built binary. You can cross-compile from source using cross build --release --target aarch64-unknown-linux-gnu -p wifi-densepose-sensing-server if you have the Rust cross-compilation toolchain set up.

Further Reading

• Architecture Decision Records - 48 ADRs covering all design decisions
• WiFi-Mat Disaster Response Guide - Search & rescue module
• Build Guide - Detailed build instructions
• RuVector - Signal intelligence crate ecosystem
• CMU DensePose From WiFi - The foundational research paper