Hi everyone,
I’ve been digging into WiFi CSI-based occupancy and motion tracking using a bunch of ESP32-S3s. In my research, I came across the RuView project.
On paper, it looks impressive—they’ve got updates as recent as 3 days ago and just dropped a pre-trained model on HuggingFace claiming it was trained on 60k samples. They’re promising full body pose estimation (DensePose) using multiple ESP32s in a multistatic mesh.
However, looking at the GitHub issues and some older threads, there’s a lot of talk about it being “vaporware” or just feeding synthetic data into a UI. Given the physics of CSI phase data, I’m a bit skeptical about achieving that kind of resolution on $9 hardware without massive noise issues.
I’ve already written my own base code to pull raw amplitude data from a single S3, but before I commit to a full 10-node deployment for this specific project, I wanted to see if anyone here has actually used it recently.
Has anyone successfully run this with a multi-node hardware setup? Does it actually deliver on the pose estimation, or is it just a well-documented concept? I’d love to hear some real-world feedback before I sink a week into the subcarrier alignment and mesh config.