Decision: the wearer's own smartphone on a chest mount is the Brain: camera, on device assistant and scene description, and the text home feature. It is out of the warning path.
Qualifying phones: iPhone models with Apple Intelligence (A17 Pro and later) on iOS 26 or later, using Apple's on device 3 billion parameter Foundation Model through the Foundation Models framework, offline and free of inference cost. Android 14 or later devices with AICore and Gemini Nano (Pixel 8 and later, Galaxy S24 and later, 6 GB RAM or more). Older phones still run the app for logging and the text home; scene description needs a qualifying device.
Why no Jetson: a modern phone already carries a neural processor that runs a multi billion parameter model on device, plus camera, GPS, cellular, speaker, microphone, and battery. A dedicated compute module duplicates all of that and adds cost, weight, heat, and a second battery.
What runs where: scene description, obstacle labeling, the assistant's routine dialog, and the text home run on device and offline. Cloud APIs (Gemini or comparable) are optional, opt in, online only, for heavy reasoning the small model cannot do. Nothing leaves the phone by default. Warning path sensor data never reaches the phone at all.
Phone battery: continuous camera plus inference is heavy, so scene description is on demand, not always on. The belt keeps working regardless of phone battery.