PaperlayerView exact revision on arXiv

Source-linked AI summary

SensWear: An Open, Modular, and AI-Ready Wearable Platform

Dariush Salami, Behzad Salami, Huseyin Yigitler

arXiv:2609.35777v1cs.HCeess.SP

TL;DR

Wearable AI/ML research needs raw, synchronized multimodal data, but commercial devices and many research platforms constrain form factor, sensing, or data pipelines. SensWear is an open, modular platform with reconfigurable hardware and raw-data interfaces; case studies show arterial PPG waveform capture and competitive heart-rate estimates while preserving raw data access.

  • Problem

    Wearable AI/ML research needs raw, synchronized multimodal data, while commercial devices fix form factor, firmware, and preprocessing and many research devices are tied to one location or sensor set.

    Source: §I · p. 1

  • Method

    SensWear combines a flexible-PCB main board, interchangeable daughter boards, and Zephyr firmware that exposes raw streams, timestamps, logging, and sensor-presence metadata across wearable forms.

    Source: §I · p. 1, §III · p. 1

  • Results

    SensWear matches Shimmer's 1.8 bpm mean absolute error and has the lowest RMSE (2.3 bpm); Oura has the highest correlation, while SensWear preserves full raw-data access.

    Source: B. Heart Rate Measurement Benchmarking · p. 3

  • Takeaways & Limitations

    Synchronized raw optical channels preserve arterial waveform morphology and support analyses beyond heart rate, including vascular aging, cardiorespiratory dynamics, stress indicators, and emerging cuffless blood-pressure models.

    Source: A. Arterial Pulse Waveform Analysis as… · pp. 2–3

    Further validation should cover motion, skin tone, dynamic activity, alternate placements, comfort, robustness, battery life, and clinical-grade ECG or equivalent references.

    Source: §VI · p. 3

Abstract

from arXiv · show

Wearable AI/ML research needs raw, synchronized, and reconfigurable multimodal data, but consumer devices are closed and many research platforms remain tied to one embodiment or sensor set. This paper presents SensWear, an open, modular, and AI-ready wearable platform that decouples embodiment, sensing, data interfaces, and learning. A compact flexible-Printed Circuit Board (PCB) main board and programmable 1.2 V to 5.5 V daughter-board interface support plug-and-play Photo- PlethysmoGraphy (PPG), touch, temperature, haptic, and LED modules across wearable form factors. Zephyr firmware pro- vides drivers, timestamping, raw streaming/logging, and sensor- presence metadata. Case studies show arterial PPG waveform capture and competitive heart-rate accuracy while preserving inspectable raw data for reproducible closed-loop experiments.

I. INTRODUCTION

Wearable AI/ML research needs raw, synchronized multimodal data, but commercial devices fix key design choices and many research devices are tied to one location or sensor set. SensWear presents an open platform that separates embodiment, sensing, data paths, and learning interfaces.

I. INTRODUCTION

  • Commercial wearables fix form factor, firmware, and preprocessing, while many research devices are tied to one body location or sensor set.These constraints limit cross-placement studies, sensor-agnostic learning, closed-loop feedback, and reproducibility.Source: §I · p. 1
  • SensWear decouples embodiment, sensing and actuation, firmware data paths, and learning interfaces in an open wearable platform.A flexible-PCB main board and interchangeable daughter boards support ring, wrist, patch, and custom forms.Source: §I · p. 1
  • The platform combines hardware, firmware, metadata, and actuation interfaces intended to keep sensing pipelines reusable across embodiments.The contribution list describes it as form-factor-agnostic and open-source.Source: §I · p. 1
  • Case studies demonstrate arterial PPG acquisition and competitive heart-rate accuracy while retaining access to raw data.The paper presents these as evidence for the platform's sensing and measurement capabilities.Source: §I · p. 1

II. RELATED WORK

Prior wearable systems often provide strong task-specific sensing or improve reproducibility within a particular embodiment. SensWear targets the gap by modularizing sensing while keeping core data and firmware interfaces stable.

II. RELATED WORK

  • Application-specific platforms provide strong sensing for particular research areas but remain tied to tasks and placements –.Examples span motion, biosignals, spectroscopy, and hearing research.Source: §II · p. 1
  • Open devices such as OmniRing and OpenEarable improve reproducibility but bind sensing to ring or ear embodiments,,.SensWear instead exposes modularity at the daughter-board boundary while keeping the main board, firmware abstractions, raw stream, metadata, and feedback interfaces stable.Source: §II · p. 1
  • Table I indicates that open systems rarely combine form-factor agnosticism, modular sensing and actuation, edge AI, and permissive reuse.Surveys also identify interoperability, energy efficiency, privacy, data management, and clinical/cloud integration as persistent wearable-AI barriers,.Source: §II · p. 1

III. DESIGN GOALS AND PRINCIPLES

SensWear is designed as reusable infrastructure, with openness, modular sensing, form-factor flexibility, AI-readiness, and bidirectional interaction as its guiding principles.

III. DESIGN GOALS AND PRINCIPLES

  • Released hardware, firmware, and software under a permissive MIT license define the platform's openness principle.The paper frames SensWear as reusable infrastructure rather than a single-purpose device.Source: §III · p. 1
  • Interchangeable daughter boards use shared I2C, GPIO, and programmable power to support sensor modularity.A compact flexible-PCB main board supports wrist, ring, patch, or custom forms.Source: §III · p. 1
  • Raw synchronized streams, local logging, metadata, and edge, mobile, or cloud processing constitute the AI-readiness principle.Haptic and visual feedback modules provide bidirectional interaction; researchers can vary placement, modality, and feedback while keeping data formats and model pipelines comparable.Source: §III · p. 1

IV. SYSTEM OVERVIEW

The system comprises a processing and communication main board, interchangeable sensing and actuation daughter boards, and a debug board for setup and serial communication.

IV. SYSTEM OVERVIEW

  • The main board handles processing, wireless communication, memory, and power management.Figure 1 depicts it as the system's central board.Source: §IV · p. 1
  • Interchangeable sensing and actuation daughter boards connect through interfaces intended to support reuse across embodiments.Table II lists Daughter-to-Main, Daughter-to-Daughter, Debug-to-Main, and Wire-to-Board interfaces.Source: §IV · p. 1
  • A debug board supports programming, reset, and UART communication with the main board.The passage identifies UART as the serial interface.Source: §IV · p. 1

A. Hardware Architecture

SensWear separates main-board computing and power from interchangeable sensing and feedback modules, using shared interfaces to support modular configurations.

A. Hardware Architecture

  • The nRF54L15 provides low-power computing and BLE, the BHI360 provides motion sensing and embedded fusion, and a 32 Mbit EEPROM supports local logging.Figure 2 depicts these components within the main-board compute/power subsystem and daughter-board sensing or feedback modules.Source: A. Hardware Architecture · p. 1
  • The platform overview shows a main board, interchangeable daughter boards, and a debug board connected through standardized interfaces for modular sensing, actuation, and development.The listed interfaces include D2M, D2D, DE2M, and W2B.Source: Figure 1 · p. 2
  • A programmable 1.2 V to 5.5 V supply, shared I2C, selected GPIO, and a D2D connector support daughter-board configurations without redesigning the main board.Current modules include touch, temperature, PPG, haptics, and LED feedback.Source: A. Hardware Architecture · p. 2

B. Firmware, Data, and AI Interfaces

Zephyr firmware standardizes device drivers and data handling, while processing can span the onboard sensors, MCU, smartphone, and cloud.

B. Firmware, Data, and AI Interfaces

  • Zephyr drivers, Devicetree mappings, and daughter-board descriptors allow hardware revisions to be registered without restructuring the firmware stack.Sensor streams use a unified system clock and can be streamed over BLE or logged locally in raw form.Source: B. Firmware, Data, and AI Interfaces · p. 2
  • Metadata records active sensors, sampling rates, and configuration parameters so downstream models can handle missing or reconfigured modalities.Processing can be distributed across the BHI360, MCU, smartphone, and cloud; the mobile application supports configuration, visualization, structured logging, edge inference, and export to cloud workflows.Source: B. Firmware, Data, and AI Interfaces · p. 2
  • The nRF54L15 supports lightweight inference and feature extraction, while Zephyr power APIs and the fuel gauge support streaming, buffered logging, and deep-sleep modes.The debug board exposes MPU pins through W2B interfaces for profiling.Source: B. Firmware, Data, and AI Interfaces · p. 2

V. CASE STUDIES

The case studies evaluate arterial pulse waveform acquisition and heart-rate estimation against commercial and research-grade references under an approved, consented protocol.

V. CASE STUDIES

  • SensWear is evaluated for arterial pulse waveform acquisition and heart-rate estimation against commercial and research-grade references.The passage does not specify the results of these evaluations.Source: §V · p. 2
  • The participant gave written informed consent for the study.Source: §V · p. 2
  • The Aalto University ethics committee approved the protocol (D/10330/03.04/2024).Source: §V · p. 2

A. Arterial Pulse Waveform Analysis as a Research Enabler

SensWear captures synchronized raw PPG channels and a filtered arterial waveform with preserved pulse morphology, supporting analyses beyond heart-rate extraction.

A. Arterial Pulse Waveform Analysis as a Research Enabler

B. Heart Rate Measurement Benchmarking

SensWear matches Shimmer’s mean absolute error and has the lowest reported RMSE, while Oura has the highest correlation. The comparison used synchronized measurements taken during rest.

B. Heart Rate Measurement Benchmarking

  • Devices were worn simultaneously during rest, with heart-rate values synchronized by measurement interval.The comparison included Shimmer PPG, Oura Ring, and an Omron cuff-based monitor used as a quasi-reference for oscillometric heart rate.Source: B. Heart Rate Measurement Benchmarking · p. 3
  • 1.8 bpm mean absolute error matches Shimmer, while SensWear’s RMSE is lowest at 2.3 bpm.Oura has the highest correlation, but SensWear maintains strong agreement while preserving full raw-data access.Source: B. Heart Rate Measurement Benchmarking · p. 3

VI. DISCUSSION AND CONCLUSION

SensWear combines a flexible-PCB main board and programmable daughter-board power with Zephyr firmware, raw synchronized streams, and sensor-presence metadata. Case studies capture arterial PPG waveforms and competitive heart-rate estimates; further validation is proposed.

VI. DISCUSSION AND CONCLUSION

  • The platform combines a flexible-PCB main board, programmable daughter-board power, shared I2C/GPIO, Zephyr firmware, raw synchronized streams, and sensor-presence metadata.The implementation repositories are available through the SensWear GitHub.Source: §VI · p. 3
  • Case studies show arterial PPG waveform capture and competitive heart-rate estimates while preserving raw-data access.Source: §VI · p. 3
  • Future validation should test motion, skin tone, activity, placement, comfort, robustness, battery life, and clinical-grade ECG-equivalent references, with reproducible hardware, configuration, placement, consent, and data-safeguard reporting.Source: §VI · p. 3
Loading 2609.35777v1…