Cognitum RuView uses WiFi CSI radio signals to detect pose, breathing, heart rate, and room occupancy through walls, with no cameras or wearables, running fully on-device.
Key Takeaways
55 KB WiFlow model (1.8M params) achieves 92.9% PCK@20 body pose accuracy from CSI data at 20 Hz across 56-192 subcarriers.
Contactless vitals: breathing at 6-30 BPM and heart rate at 40-120 BPM extracted from sub-carrier phase drift, no wearable required.
Full Rust + SIMD pipeline hits 54,000 frames/sec presence detection; 810x faster than the Python v1 baseline, running on ESP32-S3 edge hardware.
60+ Ed25519-signed WASM modules (5-30 KB each) deploy OTA via the Cognitum MCP fabric; open RVF container format allows custom Rust/C modules.
Hardware entry point is $9 per ESP32-S3 node; Cognitum Seed (presale $257) adds HNSW-indexed room fingerprints, hardware key attestation, and MCP bridge.
Hacker News Comment Review
One commenter flagged a real implementation challenge: always-on sensing systems typically waste cycles polling even when the environment is static, and RuView has not publicly addressed idle-state efficiency or interrupt-driven CSI sampling.
