RuView: $7 ESP32 Turns WiFi Signals Into Through-Wall Sensing Radar

Overview

RuView, a project trending on GitHub, is built on a counterintuitive physical fact: the WiFi signal from your router is already reflecting off your body. Every breath-induced chest movement and every heartbeat subtly perturbs those 2.4GHz electromagnetic waves. The question was never "can we sense it," but "how do we decode the sensed signal."

RuView's answer is surprisingly simple: a $7 ESP32-S3 development board, open-source firmware, and WiFi CSI (Channel State Information) fed into a small locally-running model. The output: 17-body-point pose estimation, breathing rate, heart rate, fall detection, and sleep staging.

A router plus a thumb-sized board. No camera, no wearables. It can detect, through walls, which room you're in, whether you're walking or sitting, your breathing rate, your heart rate—even whether you experience sleep apnea.

Physics: WiFi Signals Are Already "Radar"

Radar works by transmitting electromagnetic waves, receiving reflections, and deriving target position and velocity from time delays and frequency shifts. WiFi routers emit thousands of electromagnetic pulses per second, and your body—mostly water—reflects some of them. Conventional WiFi chips only care about whether the signal decodes into 0s and 1s, discarding reflections as noise.

CSI is that discarded portion: it records the amplitude and phase of every subcarrier at every moment—essentially a fine-grained snapshot of "how the signal was distorted in space."

• A person walking across a room draws a clear trajectory in the CSI waveform.
• A person sitting still while breathing produces periodic micro-fluctuations at 12–20 cycles/minute.
• A sleeping person shows the slow breathing cycle plus high-frequency micro-fluctuations at 60–100 cycles/minute corresponding to heart rate—visible as two distinct peaks after an FFT.

RuView's pipeline: ESP32 captures CSI → preprocessing and denoising → trained model → semantic events ("someone is sleeping," "possible fall," "bathroom occupied").

Engineering: 21 Entities per Node

RuView is not a paper—it's a working system. Each ESP32 node outputs 21 entities: 11 raw signals (CSI amplitude, phase, SNR, etc.) plus 10 inferred semantic states:

• someone-sleeping
• possible-distress
• room-active
• elderly-inactivity-anomaly
• meeting-in-progress
• bathroom-occupied
• fall-risk-elevated
• bed-exit
• no-movement
• multi-room-transition

These are not low-level events like "motion detected" but semantically-inferred results. "Elderly inactivity anomaly" isn't "no one is moving"—it's "someone should normally be active at this time, but no micro-movement has been detected for 2 consecutive hours." "Fall risk elevated" isn't "a fall happened"—it's "gait CSI patterns resemble the pattern 30 seconds before historical falls."

This is a three-layer architecture of sensing → reasoning → semantics, isomorphic to computer vision's pixels → features → object recognition. The difference: RuView's "pixels" are CSI time series, not RGB.

Privacy: Surveillance Without a Lens

The camera's problem isn't "it can see"—it's that what it captures can be replayed, stolen, and used to train facial recognition models. RuView has no lens. It outputs data like "someone in the room, breathing 16/min, posture: lying"—data worthless to face recognition models but enormously valuable for "is the elderly person living alone safe."

A key distinction: the privacy cost of surveillance depends not on what can be sensed, but on whether the sensed signal can be reverse-engineered into personal biometric identity. CSI is a highly semantic abstraction: it tells you "a person is breathing" but not "who that person is."

That said, CSI is not absolutely anonymous. Research has shown gait CSI patterns can be used for identification—each person's walking rhythm, stride, and weight shift is unique. RuView's documentation explicitly acknowledges this, attaching privacy policies, uncertainty estimates, provenance, and witness records to each sensing event. It's a "governance-first" engineering attitude rather than "ship first, ask later."

MetaHarness: An Honesty Check for AI Claims

A less conspicuous but important part of the RuView repository is MetaHarness: an AI operator that guides humans and AI agents through deployment, calibration, training, and verification, and performs "honesty checks" on all sensing claims.

Why? WiFi sensing has a chronic problem: 95% accuracy in papers can drop to 60% in your living room—different router placement, wall materials, other reflectors, models trained on lab data. Without on-site calibration and verification, claimed accuracy is a bad check.

MetaHarness requires every claim to carry source citations, deterministic verification, and on-site data validation. Claims are not capability; claims plus evidence are capability. RuView's author built this into the system design rather than leaving it to users.

Cross-Domain Analogy: Electric Fish and Bats

RuView's physics has direct biological parallels. Weakly electric fish navigate and hunt in the murky Amazon by emitting electrical pulses and sensing reflected field changes. Bats use ultrasonic echolocation. Neither relies on lens-based vision—they build environment models from actively emitted signals plus fine-grained reflection analysis.

RuView is the artificial version of this strategy. More interestingly, its "unified RF world model" attempts to unify WiFi CSI, millimeter-wave radar, UWB, and cellular sensing into one scene model—mirroring multimodal sensory fusion in biology: owls locate with vision + hearing, sharks with smell + electroreception. Single modalities have blind spots; multimodal fusion is robust.

Why Now

WiFi sensing isn't new—MIT's 2013 WiTrack paper demonstrated through-wall human tracking with WiFi. But the field was stuck on three bottlenecks:

1. Hardware: Getting CSI required dedicated SDRs or modified drivers. The ESP32-S3 is the first cheap consumer chip that directly outputs CSI. 2. Models: Traditional methods relied on hand-crafted features with poor generalization. Transformers and LoRA make small-dataset training possible. 3. Ecosystem: No standardized semantic output layer existed. RuView's 21-entity model plus Home Assistant/Matter/HomeKit integration is the first time WiFi sensing has been plugged into standard smart home protocols.

With all three bottlenecks loosening simultaneously, $7 hardware + open-source firmware + pretrained models + smart home integration pushes through-wall sensing from the lab into purchasable DIY kits.

A Deeper Observation

RuView illustrates a universal cross-domain principle: the best sensor is one you've already deployed but aren't using.

WiFi routers were already transmitting; the signals were already being reflected. For decades we used only the "does the signal decode into 0s and 1s" layer, discarding finer-grained CSI as noise. RuView's contribution is not "inventing a new sensing method"—it's "recovering the information that was thrown away."

The corollary: you probably have many "free sensors" being wasted around you. Bluetooth broadcast packets, ambient light sensors, microphone noise floors, accelerometer micro-vibrations—all contain hidden information that nobody decodes. RuView provides a reusable engineering template for "decoding already-existing signals."

$7. One ESP32. One open-source repo. Watching you breathe through walls, without a lens. Not science fiction—2026's GitHub trending.

---

Project: https://github.com/ruvnet/RuView Model repo: https://huggingface.co/ruvnet/wifi-densepose-pretrained Author blog: https://ruview.blog/wifi-csi-sensing-open-source