What iMuninn is

iMuninn turns an iPhone into a sensor node on your network: camera (H.264), IMU, ARKit 6-DoF pose with per-frame intrinsics and LiDAR depth, sent over the LAN (Wi‑Fi or USB) with a Protobuf protocol. The phone is the server; receivers connect to it, several at once if you like.

Three ways to receive

Way	Good for	Start here
Python package imuninn	Scripts, datasets, robot stacks, Rerun visualisation, MCAP recording, an MCP server for agents	Install → Quick start
Web viewer	A quick look at the picture and the plots: Safari opens the online page; Chrome uses imuninn web locally	Web viewer
C++ library	Real-time control programs that link it directly	C++ client

All three receive the same messages (imuninn.proto); only the transport differs: Python / C++ use TCP + UDP, browsers use WebSocket.

What you get

Stream	Content	Typical rate
ar_pose	ARKit pose (right-handed, Y up, camera looks down −Z) + camera intrinsics + tracking state	60 Hz
imu	Acceleration + angular rate (CoreMotion fused, or raw sensors)	100–200 Hz
rgb	H.264 video, intrinsics and rotation_deg on every frame	30–60 fps
depth	LiDAR depth 256×192, millimetres + confidence, LZ4	60 Hz (Pro models)
health	Thermal state, battery, phone-side frame rates, encoder drops	0.2–5 Hz

Every parameter (rates, resolution, lens, exposure / white balance / focus, bitrate, black-screen mode…) can be changed remotely; the phone validates and reports applied / clamped / rejected.

Measured latency (iPhone 17 Pro, USB)

Stream	Median one-way latency
imu	2.6 ms
ar_pose	39 ms (mostly ARKit’s own pipeline)
rgb 1920×1440@60	49 ms
depth	44 ms

Over Wi‑Fi the control round trip is about 8 ms; a 3-hour Wi‑Fi run dropped 0.04% of video frames with the phone’s thermal state staying nominal (on charger).

Next

1. Install the app and connect (Chinese for now)
2. Pick a receiver and start