Community Discussion · Policy

The barrier for embedded AI lies in iteration, not algorithms

Bili GeAug 222026/08/22 309 views

Just saw that No Starch Press made Chapter 9 of Embedded AI available for free download. It covers sensor machine learning, basically teaching you how to move models from the cloud to MCUs, running them right next to the sensors. It's a niche direction, but several projects I've been looking at recently can't avoid it.

Policy

3 replies

Ctrl + Enter to reply

Production Line VeteranAug 23

What jiang_yuyan said about test coverage might be enough for consumer electronics, but on industrial production lines, equipment updates have to pass line validation. One failed OTA could stop the line for hours. When you do the ROI math, it might actually be worse than not updating at all.

Ming Ming Bu Gui FanAug 22

Toolchains and OTA are the basics, but what I care about more is test coverage. Updating on-device models can't just rely on pushing binaries via OTA; you need an automated regression test suite to verify things on the device, otherwise every update feels like a gamble.

Lü WenboAug 22

True... I used to tweak three versions a day using cloud models, but doing it this way on-device isn't realistic. Once things are deployed, changing parameters means waiting for processes to complete... It's tough.