Why this matters
Wi‑Fi CSI motion sensing turns ubiquitous 2.4 GHz signals into a privacy‑focused presence sensor that requires no cameras or microphones and can run on a ~€10 ESP32. That shift — using radio multipath variations rather than visual or audio data — makes reliable, always‑on occupancy signals practical for smart homes and low‑cost monitoring, while keeping sensing local and low‑bandwidth.
What Sets It Apart
- Automatic subcarrier selection (NBVI) and lightweight signal pipeline — so what: reduces per‑site tuning and yields high detection performance in many typical home layouts.
- On‑device ML detector (experimental) with zero calibration — so what: enables edge inference without cloud dependency and simplifies deployment for non‑technical users.
- Native ESPHome + Home Assistant integration — so what: sensors are auto‑discovered and expose both binary motion and a continuous movement score for automation and analytics.
- Two‑platform strategy (ESPectre for production, Micro‑ESPectre for R&D) — so what: hobbyists get a plug‑and‑play component while researchers can experiment with Python tooling and algorithm development.
Who it's for — and tradeoffs
Great fit if you run Home Assistant, want a privacy‑preserving occupancy sensor, or are building low‑cost ambient monitoring (elderly care, energy saving, basic security). It is also useful for researchers who need a fast CSI prototyping stack via Micro‑ESPectre.
Look elsewhere if you need person identification, fine‑grained activity classification, or guaranteed performance across very noisy/metallic environments — accuracy is environment dependent and the built‑in detectors report generic motion (not person vs pet). Also note ethical and legal responsibilities: the system can detect presence without audio/visual data, so obtain consent and follow local privacy regulations when deploying.
Practical signals and maturity
The repository couples production‑oriented firmware with documentation and tuning guides; it exposes configuration through YAML via ESPHome and ships algorithm docs (ALGORITHMS.md) and performance notes (PERFORMANCE.md). The project is widely adopted in the smart‑home community and provides both a ready‑to‑use Home Assistant component and an extensible research path for those who want to train or evaluate ML detectors.
