Track movement and vitals through walls using WiFi signals
RuView turns in-home WiFi signals into an unattended monitoring system that detects presence, respiration, and heart rate. It tracks activity within a space using existing routers and low-cost ESP32 boards, creating a privacy-preserving monitoring environment without video data. The solution is based on analyzing WiFi Channel State Information (CSI) to convert physical phenomena into data.
Spatial intelligence from WiFi signals
RuView measures how WiFi radio waves filling a room are distorted by people. ESP32 sensors capture these subtle changes and convert them into Channel State Information (CSI), which is used to determine presence, movement, and respiration cycles. No cameras or wearable devices are required, and the system works through walls or in dark areas. It scans six WiFi channels and supports multi-frequency mesh scanning that utilizes neighboring router signals as free radar illumination. All measurements are secured for integrity through an Ed25519 encrypted witness chain.
Real-time vital signs and pose estimation
Respiration rate and heart rate are measured in real time without contact, maintaining accuracy even during sleep. The system includes pose estimation for 17 major body landmarks, though reliability varies by current state. A pre-trained model published on Hugging Face shows 82.3% holdout accuracy and can be processed in microseconds on a Raspberry Pi with a size of 8KB when 4-bit quantized. In contrast, the on-device pose model running directly on ESP32 is still in its early stages, recording a PCK@20 of 3.0% against a target accuracy of 35%, placing it in a pre-practical phase. On the MM-Fi benchmark, it achieves 82.69% accuracy, surpassing existing SOTA, but this is based on research datasets and differs from actual single ESP32 environments.
Smart home ecosystem integration and local automation
Native integration is supported with Home Assistant, Apple Home, Google Home, and Amazon Alexa. For Home Assistant, an MQTT publisher exposes 21 entities (11 raw signals and 10 inferred semantic states), along with three starter blueprints. For Apple Home and other ecosystems, pairing via a Matter bridge allows querying room-by-room presence and vital signs through voice assistants. The HOMECORE module provides state management, history, automation, signed Wasm plugins, and HomeKit support, enabling a complete local automation loop without the cloud. An AI operator (MetaHarness) guides onboarding, calibration, learning, and verification processes, and integration with external AI tools like ChatGPT is possible via an MCP server.
Hardware configuration and installation paths
The basic configuration consists of an ESP32-S3 board (approximately $9) and a Cognitum Seed, with a total BOM cost of about $140. Configuring only the ESP32 mesh costs approximately $54 and allows use of core detection functions, while pipeline verification is also possible using research NICs or simulators. A Docker image allows evaluation with simulation data without hardware, and development environments can be set up using the Python package (ruview) or the npm package (@ruvnet/ruview). The ESP32-C6 board supports WiFi 6 and 802.15.4 research detection, and battery-based node operation is possible through low-power mode and TWT settings. All edge modules are distributed as signed binaries, and devices update the catalog via OTA and verify integrity before installation.
Limitations and status to check before adoption
While this system is fully cloud-independent, some features are still in the research stage or limited. Specifically, pose estimation in a single-antenna ESP32 environment has not yet reached a reliable level, which is clearly stated in the README. Among the published models, the model.safetensors file may be rejected by some loaders due to header issues, so using csi-embed-v2.safetensors or the JSONL format is safer. When evaluating with Docker without hardware, only coarse presence detection based on RSSI is possible, and precise CSI-based features require an ESP32-S3 or a research NIC. Since the license and long-term maintenance policy are not specified in the README, legal review is necessary before commercial distribution.
