Is this a real and usable project？

[AfricanCh]

I haven't tried it yet, which requires some cost. I have seen many issues with reported issues being closed as' unplanned ', which makes me cautious. Please forgive me if I have offended you

[p20061]

我也感觉很奇幻，那这么来说蓝牙应该也可以，只是长的电磁波

[oki-zon3]

Philips, with its WiZ smart Wi-Fi products, uses a technology similar to Spacesense, which is Wi-Fi-based motion detection. It works by having at least two products (e.g., smart Wi-Fi lamps) communicate continuously in the same room. When an object moves in the room, a small change occurs in the communication process between the devices. This change is then analyzed by the device firmware as motion detection.

[36mimu36]

Yes, it works — tested with real hardware

I deployed the Python sensing pipeline on real hardware and can confirm it detects human presence and motion.

Hardware

• 2x ESP32-S3-DevKitC (about $10 each)
• 1x Raspberry Pi (running Debian 13)
• Standard home WiFi router

What works

• Presence detection: Correctly identifies person in room (PRESENT_STILL, 80% confidence)
• Motion detection: Motion Band power rises when person moves (0.33 measured)
• Multi-node: Two ESP32 nodes streaming CSI simultaneously, cross-receiver agreement boosts confidence
• Real-time UI: Web UI at http://<pi>:8080 shows live signal field, RSSI, classification

Setup steps (bare metal, no Docker)

1. Flash ESP32-S3 with the firmware in firmware/esp32-csi-node/ using ESP-IDF v5.2
2. Set WiFi SSID/password and target IP in sdkconfig.defaults
3. On Pi: install Python deps (numpy, scipy, websockets, pydantic), run ws_server.py
4. ESP32 auto-streams ADR-018 binary CSI frames over UDP to the Pi

Total cost: ~$30 (2x ESP32-S3) + a Pi you may already have.

Caveats

• Breathing detection (0.1–0.5 Hz) does not work yet with the Python pipeline — needs per-subcarrier phase tracking
• The default mean_amplitude signal is noisy; I switched to subcarrier spread (np.std(amplitudes)) for better results (see my comment on Amplitude-based detection is fragile in non-static environments #51)
• Full DensePose (body mesh) requires trained neural network models not yet available — current capability is presence + motion level classification

It's not magic — it's WiFi CSI (Channel State Information), the same principle behind Philips WiZ Spacesense mentioned above. This project provides an open-source implementation.

[ruvnet]

I'm no longer actively building the Python version. I should note almost everyone who said it doesn't work didn't actually use CSI compatible hardware.

Thanks for the confirmation.

[rezzorix]

almost everyone who said it doesn't work didn't actually use CSI compatible hardware.

May i suggest you make it more prominent in the initial description or right below the header in BOLD that certain hardware is required?

Bonus: maybe provide an example list of hardware that is working.

[maisSistemasDev]

I don't think it works. I removed the Wi-Fi adapter and it continued showing the hardware and the skeleton.

[perdakovich]

нет это ии слоп с накручеными звёздами

[lplzyp]

Yes, it works — tested with real hardware

I deployed the Python sensing pipeline on real hardware and can confirm it detects human presence and motion.

Hardware

• 2x ESP32-S3-DevKitC (about $10 each)
• 1x Raspberry Pi (running Debian 13)
• Standard home WiFi router

What works

• Presence detection: Correctly identifies person in room (PRESENT_STILL, 80% confidence)
• Motion detection: Motion Band power rises when person moves (0.33 measured)
• Multi-node: Two ESP32 nodes streaming CSI simultaneously, cross-receiver agreement boosts confidence
• Real-time UI: Web UI at http://<pi>:8080 shows live signal field, RSSI, classification

Setup steps (bare metal, no Docker)

1. Flash ESP32-S3 with the firmware in firmware/esp32-csi-node/ using ESP-IDF v5.2
2. Set WiFi SSID/password and target IP in sdkconfig.defaults
3. On Pi: install Python deps (numpy, scipy, websockets, pydantic), run ws_server.py
4. ESP32 auto-streams ADR-018 binary CSI frames over UDP to the Pi

Total cost: ~$30 (2x ESP32-S3) + a Pi you may already have.

Caveats

• Breathing detection (0.1–0.5 Hz) does not work yet with the Python pipeline — needs per-subcarrier phase tracking
• The default mean_amplitude signal is noisy; I switched to subcarrier spread (np.std(amplitudes)) for better results (see my comment on Amplitude-based detection is fragile in non-static environments #51)
• Full DensePose (body mesh) requires trained neural network models not yet available — current capability is presence + motion level classification

It's not magic — it's WiFi CSI (Channel State Information), the same principle behind Philips WiZ Spacesense mentioned above. This project provides an open-source implementation.

Yes, it works — tested with real hardware

I deployed the Python sensing pipeline on real hardware and can confirm it detects human presence and motion.

Hardware

• 2x ESP32-S3-DevKitC (about $10 each)
• 1x Raspberry Pi (running Debian 13)
• Standard home WiFi router

What works

• Presence detection: Correctly identifies person in room (PRESENT_STILL, 80% confidence)
• Motion detection: Motion Band power rises when person moves (0.33 measured)
• Multi-node: Two ESP32 nodes streaming CSI simultaneously, cross-receiver agreement boosts confidence
• Real-time UI: Web UI at http://<pi>:8080 shows live signal field, RSSI, classification

Setup steps (bare metal, no Docker)

1. Flash ESP32-S3 with the firmware in firmware/esp32-csi-node/ using ESP-IDF v5.2
2. Set WiFi SSID/password and target IP in sdkconfig.defaults
3. On Pi: install Python deps (numpy, scipy, websockets, pydantic), run ws_server.py
4. ESP32 auto-streams ADR-018 binary CSI frames over UDP to the Pi

Total cost: ~$30 (2x ESP32-S3) + a Pi you may already have.

Caveats

• Breathing detection (0.1–0.5 Hz) does not work yet with the Python pipeline — needs per-subcarrier phase tracking
• The default mean_amplitude signal is noisy; I switched to subcarrier spread (np.std(amplitudes)) for better results (see my comment on Amplitude-based detection is fragile in non-static environments #51)
• Full DensePose (body mesh) requires trained neural network models not yet available — current capability is presence + motion level classification

It's not magic — it's WiFi CSI (Channel State Information), the same principle behind Philips WiZ Spacesense mentioned above. This project provides an open-source implementation.

Hello, I don't have the equipment on hand right now, so I'd like to trouble you. I have a general understanding of how CSI works, but I still have some questions about the areas and functions this project claims to be capable of. For example, regarding human pose recognition, you mentioned that it requires a pre-trained model. Is this model extremely closely related to the environment? For instance, if I use a trained model in the current indoor environment and it can recognize human poses, but I change the tables, chairs, sofas, or even change the environment altogether, will the system fail? Or will the error increase?

[akhil-rana]

Can someone just attach a video capture recording, if all this is real, put it in the readme for proof and problem solved.
And like others have pointed out, when this project works, it keeps working even when they disconnected the adapter.

But no, that won't happen and all of us will keep on talking to openclaw bots in GH issues and our comments/issues will keep on getting deleted like they never existed.

This the github version of karma farming and it's all so pointless.

[lplzyp]

这是来自QQ邮箱的假期自动回复邮件。您好，我最近正在休假中，无法亲自回复您的邮件。我将在假期结束后，尽快给您回复。

[grumpybozo]

I'm no longer actively building the Python version. I should note almost everyone who said it doesn't work didn't actually use CSI compatible hardware.

It would be unwise to do so until you have some way of actually using data instead of simply generating it randomly.

Thanks for the confirmation.

Yes, it has provided another user whose GH contributions are suspect. Along with you.

[zwzheng45]

I deployed the Python sensing pipeline on real hardware and can confirm it detects human presence and motion.

The python pipeline will not work with ANY hardware.

RuView/v1/src/core/router_interface.py

Lines 152 to 169 in 407b46b

	async def _collect_real_csi_data(self) -> Optional[np.ndarray]:
	"""Collect real CSI data from the router.
	Raises:
	RuntimeError: Always in the current state, because real CSI
	data collection requires hardware setup that has not been
	configured. This method must never silently return random
	or placeholder data.
	"""
	raise RuntimeError(
	f"Real CSI data collection from router '{self.router_id}' requires "
	"hardware setup that is not configured. You must: "
	"(1) install CSI-capable firmware (e.g., Atheros CSI Tool, Nexmon CSI) on the router, "
	"(2) configure the SSH connection to the router, and "
	"(3) implement the CSI extraction command for your specific firmware. "
	"For development/testing, use mock_mode=True. "
	"See docs/hardware-setup.md for complete setup instructions."
	)

It always raise this RuntimeError and tell you to see the docs/hardware-setup.md for setup instructions, which simply doesn't exist. What a joke!

It's very clear that this is just a piece of AI generated junk, we should all stop wasting time on here.

[maisSistemasDev]

I don't think it works. I removed the Wi-Fi adapter and it continued showing the hardware and the skeleton.

Sorry. I hadn't checked the model of my Wi-Fi adapter. I'm adapting the code to use with the AC1300.