Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CycloWatt: An Affordable, TinyML-enhanced IoT Device Revolutionizing Cycling Power Metrics

Victor Luder, Sizhen Bian, Michele Magno

TL;DR

A cutting-edge Internet of Things device that seamlessly integrates force signals with inertial sensor data while leveraging the power of edge machine learning techniques is introduced, marking a significant milestone in developing cost-effective and accurate cycling power meters.

Abstract

Cycling power measurement is an indispensable metric with profound implications for cyclists' performance and fitness levels. It empowers riders with real-time feedback, supports precise training regimen planning, mitigates injury risks, and enhances muscular development. Despite these advantages, the widespread adoption of cycling power meters has been hampered by their prohibitive cost and deployment complexity. This paper pioneers a groundbreaking approach to power measurement in cycling, prioritizing affordability and user-friendliness. To achieve this goal, we introduce a cutting-edge Internet of Things (IoT) device that seamlessly integrates force signals with inertial sensor data while leveraging the power of edge machine learning techniques. In-field experimental evaluations demonstrate that our prototype can estimate power with remarkable accuracy, boasting a Mean Absolute Error (MAE) of only 12.29 Watts (4.1\%). Notably, our design emphasizes energy efficiency, operating in a low-power mode that consumes a mere 50 milliwatts and offers an exceptional battery life of up to 25.8 hours in always-on active mode. With an ultra-low latency of 4.33 milliseconds for data processing and inference, our system ensures real-time power estimation during cycling activities. Incorporating IoT concepts and devices, this paper marks a significant milestone in developing cost-effective and accurate cycling power meters.

CycloWatt: An Affordable, TinyML-enhanced IoT Device Revolutionizing Cycling Power Metrics

TL;DR

A cutting-edge Internet of Things device that seamlessly integrates force signals with inertial sensor data while leveraging the power of edge machine learning techniques is introduced, marking a significant milestone in developing cost-effective and accurate cycling power meters.

Abstract

Cycling power measurement is an indispensable metric with profound implications for cyclists' performance and fitness levels. It empowers riders with real-time feedback, supports precise training regimen planning, mitigates injury risks, and enhances muscular development. Despite these advantages, the widespread adoption of cycling power meters has been hampered by their prohibitive cost and deployment complexity. This paper pioneers a groundbreaking approach to power measurement in cycling, prioritizing affordability and user-friendliness. To achieve this goal, we introduce a cutting-edge Internet of Things (IoT) device that seamlessly integrates force signals with inertial sensor data while leveraging the power of edge machine learning techniques. In-field experimental evaluations demonstrate that our prototype can estimate power with remarkable accuracy, boasting a Mean Absolute Error (MAE) of only 12.29 Watts (4.1\%). Notably, our design emphasizes energy efficiency, operating in a low-power mode that consumes a mere 50 milliwatts and offers an exceptional battery life of up to 25.8 hours in always-on active mode. With an ultra-low latency of 4.33 milliseconds for data processing and inference, our system ensures real-time power estimation during cycling activities. Incorporating IoT concepts and devices, this paper marks a significant milestone in developing cost-effective and accurate cycling power meters.
Paper Structure (11 sections, 5 figures, 4 tables)

This paper contains 11 sections, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. System Overview
  4. Hardware
  5. Data Set and Data Collection Set-Up
  6. Data Processing
  7. Proposed TinyML Model
  8. Experimental Evaluation
  9. Prototype Testing
  10. Power Consumption and Latency
  11. Conclusion

Figures (5)

  • Figure 1: Full disassembled hardware setup and final assembled device (Left to right: pedal to force sensor interface, force sensor, custom cleat, electronics and assembled power meter).
  • Figure 2: Data collection setup with reference measurement. Illustration of the parallel collection and alignment of reference data and measured data.
  • Figure 3: Histogram of the collected training data, showing the distribution and balance of the training data.
  • Figure 4: Results of the three indoor test rides, using the unquantized model.This compares the reference power measurement (black) with the estimated power (red).
  • Figure 5: Results of the 11.3 kilometer outdoor test ride, with reference power (black) and estimated power (red).