ADR-041: Edge Intelligence — 60 WASM Modules for On-Device WiFi Sensing

[ruvnet]

What This Is

Your WiFi router already fills every room with radio waves. When someone walks through, breathes, or even just stands still, those waves bounce differently. WiFi DensePose reads those changes to understand what is happening in a room — without cameras.

These 60 edge modules are small programs that run directly on a $8 ESP32 chip. They analyze WiFi signals and make decisions instantly, right on the device — no internet connection needed, no cloud fees, no lag.

Each module is a tiny file (5-30 KB) that you upload to the chip wirelessly. It starts working immediately.

---

What Can It Actually Do?

Health Monitoring (no wearables needed)

Put an ESP32 in a bedroom or hospital room. It watches WiFi reflections off a person's chest and body to detect:

Module	What it does in plain English
Sleep Apnea	Notices when someone stops breathing during sleep for more than 10 seconds. Counts how many times per hour.
Heart Rhythm	Watches for heart rate that is too fast (>100), too slow (<50), or irregular.
Breathing Trouble	Catches rapid shallow breathing, labored breathing, or the waxing-and-waning pattern seen in heart failure.
Walking Analysis	Detects shuffling gait, unsteady walking, and estimates fall risk on a 0-100 scale.
Seizure Detection	Recognizes the rhythmic shaking of a seizure and distinguishes it from a normal fall.
Vital Trends	Tracks breathing and heart rate over hours and days. Alerts when someone is slowly getting worse.

Security (no cameras, works in the dark)

Mount ESP32 nodes around a building. They sense movement through walls, in total darkness, without any visible equipment.

Module	What it does in plain English
Intrusion Detection	Knows the difference between a person walking in and a pet or air conditioning.
Perimeter Watch	Tells you which direction someone is moving — approaching or leaving — across up to 4 zones.
Metal Detection	Experimental: spots large metallic objects (weapons) from how they reflect WiFi differently than a human body.
Tailgating	Catches when two people slip through a secured door together instead of one.
Loitering	Alerts when someone stays in one spot for too long (you set the time limit).
Panic/Distress	Recognizes frantic, erratic movement — running, struggling, or fleeing.

Smart Buildings (save energy, manage space)

Use existing office WiFi or add a few ESP32 nodes. Know which rooms are actually being used.

Module	What it does in plain English
Zone Occupancy	Shows which parts of a room are occupied on a zone-by-zone grid.
HVAC Control	Tells heating/cooling whether anyone is in the room — with smart delays so it does not flip-flop when someone goes to the bathroom.
Lighting Control	Turns lights on when someone enters, dims after 10 quiet minutes, off when they leave. Per-zone.
Elevator Count	Counts how many people are in the elevator and warns if it is overloaded.
Meeting Rooms	Knows if a meeting is actually happening or if someone just ducked in for 30 seconds. Tracks start, end, headcount.
Energy Audit	Builds a weekly schedule of when rooms are used and when they are empty — so you can turn things off during dead hours.

Retail & Restaurants (understand customers without cameras)

No GDPR issues, no consent forms, no video storage. Just WiFi signals.

Module	What it does in plain English
Queue Length	Counts how many people are waiting in line and estimates wait time.
Dwell Heatmap	Shows which areas of a store get the most foot traffic and for how long.
Customer Flow	Counts people entering vs leaving, tracks net occupancy.
Table Turnover	For restaurants: tracks when tables are seated, occupied, and cleared. Calculates turnover rate.
Shelf Engagement	Detects when a customer stops to look at products: quick browse vs deep consideration.

Industrial Safety (works through dust, smoke, shelving)

WiFi passes through obstacles where cameras fail. Critical for worker safety.

Module	What it does in plain English
Forklift Proximity	Warns when a person is too close to a forklift or robot — even around blind corners.
Confined Space	Monitors workers in tanks or manholes: confirms they are breathing, alerts if they stop moving.
Clean Room	Enforces maximum occupancy and flags turbulent movement that could contaminate the environment.
Livestock	Monitors animals in barns: detects illness (abnormal stillness), labored breathing, escape.
Structural Vibration	Detects building vibration, earthquake signals, and equipment resonance from WiFi phase stability.

Signal Processing (runs on the chip)

Advanced algorithms that clean and enhance the WiFi signals before other modules use them.

Module	What it does in plain English
Flash Attention	Focuses processing on the WiFi channels that carry the most useful information.
Coherence Gate	Throws away bad/noisy signal frames so they do not corrupt results.
Temporal Compress	Compresses old data to save memory: recent data kept at full quality, older data reduced.
Sparse Recovery	Fills in missing data points when some WiFi channels drop out.
Person Match	Keeps track of which person is which across consecutive frames so identities do not swap.
Optimal Transport	Measures how much the WiFi environment has changed (new person entered, furniture moved).

On-Device Learning (the chip teaches itself)

No cloud training. The ESP32 learns from what it sees.

Module	What it does in plain English
Gesture Learning	You teach it custom gestures by repeating them 3 times. It remembers up to 16.
Anomaly Detection	Learns what "normal" looks like and flags when something unusual happens.
Self-Tuning	Automatically adjusts its own sensitivity thresholds — gets better over time.
Lifelong Learning	Learns new things without forgetting old things (a hard problem in AI).

Spatial & Temporal Intelligence

Module	What it does in plain English
Influence Mapping	Figures out which room zones affect each other (e.g., hallway traffic predicts lobby activity).
Fast Search	Quickly matches current WiFi patterns against known room states.
Zone Tracking	Brain-inspired zone tracking using simulated neurons that adapt over time.
Routine Detection	Learns daily patterns and alerts when someone deviates (e.g., "Mom usually moves by 8am").
Safety Rules	Checks safety conditions continuously: "if no movement for 60 seconds, alert."
Autonomous Planning	The chip decides which modules to activate based on what is happening.

Security & Tamper Protection

Module	What it does in plain English
Signal Shield	Detects if someone is trying to replay, jam, or inject false WiFi signals.
Behavior Profiler	Builds a profile of normal activity and flags anomalies using statistical scoring.

Research & Exotic

Module	What it does in plain English
Sleep Staging	Classifies sleep phases (light, deep, REM) from breathing and movement patterns.
Emotion Sensing	Estimates stress vs calm from breathing patterns, heart rate, and fidgeting.
Sign Language	Recognizes hand/arm gestures — a step toward WiFi-based sign language reading.
Music Conductor	Tracks conducting gestures for tempo, dynamics, and beat position — an air instrument.
Plant Monitor	Detects plant growth, circadian leaf movement, and wilting over days (needs empty room).
Ghost Hunter	Detects unexplained disturbances in empty rooms — actually useful for finding gas leaks, hidden intruders, or pests.
Rain Detection	Senses rain hitting the roof from vibrations that travel through the building.
Breathing Sync	Detects when multiple people's breathing synchronizes — meditation quality, couple sleep, crowd coherence.
Time Patterns	Finds hidden periodic patterns in activity that repeat at unexpected intervals.
Location Embedding	Maps room locations in a mathematical space that naturally represents hierarchical relationships.

---

How to Use

Prerequisites

• Rust toolchain with wasm32-unknown-unknown target
• ESP32-S3 with WASM3 firmware (ADR-040)

Build and test

cd rust-port/wifi-densepose-rs/crates/wifi-densepose-wasm-edge
cargo check --features std          # verify it compiles
cargo test --features std           # run all tests
cargo build --target wasm32-unknown-unknown --release  # build for ESP32

Upload to ESP32

python scripts/wasm_upload.py --port COM7 --module target/wasm32-unknown-unknown/release/module_name.wasm

---

Architecture

Each module talks to the ESP32 through 12 simple functions:

Function	What it gives you
csi_phase(i)	WiFi signal phase angle for channel i
csi_amplitude(i)	WiFi signal strength for channel i
csi_variance(i)	How much channel i fluctuates
csi_mean_amplitude()	Average signal strength
csi_motion_energy()	Overall movement level
csi_n_persons()	Estimated number of people
csi_is_present()	Is anyone there? (yes/no)
csi_breathing_bpm()	Breathing rate
csi_heart_bpm()	Heart rate
emit_event(id, value)	Send a detection result to the host

---

Security

All modules have been security audited. No unbounded memory, no unsafe allocation, all inputs validated. See the full audit report.

Related

• ADR-039 — How edge processing tiers work
• ADR-040 — The WASM runtime design
• ADR-041 — Full specification for all 60 modules

[Alessandropantoja01]

Gostei do projeto, good job 👏👏👏