Porting an AI Powered Wearable Health Monitor to Zephyr on Open Hardware
summary
Closed-source single-threaded wearable migrates to Zephyr RTOS on QuickFeather EOS S3 with drivers added for Wishbone/I2C and the MAX30102 SPO2 sensor; bootloader reflashes require J-Link after OpenOCD failures, and the eFPGA proves too small for ML with two 32×32 multipliers so inference runs on the Cortex‑M4 using TensorFlow Lite int8 quantized models. Preprocessing yields a 6‑input feature vector, prototype on a breadboard works, and measured error is modestly high at MAE systolic 16.5 mmHg and diastolic 8.3.
AI-written summary · about
chapters (9)
- 0:00Project introduction
- 3:08Original design and hardware
- 6:10Transition to Zephyr and EOS S3
- 8:57Overcoming hardware hurdles
- 13:04Developing drivers for Zephyr
- 19:10Driver integration and testing
- 24:44Deploying machine learning
- 28:17Blood pressure prediction
- 34:07Future work and conclusions
related talks
07mLightning Talk: Zephyr Portability with an AI Application on Very Different MCUs
14mKeynote: Leveraging Zephyr and ML to Bring Smart Devices to Market, Faster
01h 19mMini-conference: RISC-V Support in Zephyr
39mFrom a Weekend Hack, to a Future-Proof IoT Device
10mLightning Talk: Debugging Edge AI on Zephyr and Lessons Learned
35mEnd-to-end IoT Development with Zephyr
01h 05mBoF: Unlocking Software Defined Power Electronics with Zephyr: Concepts...
01h 16mOpening Keynote