A Multi-Modal Dataset for Ground Reaction Force Estimation Using Consumer Wearable Sensors
A recent study published on arXiv (arXiv:2603.28784v1) presents a comprehensive multi-modal dataset aimed at estimating vertical ground reaction force (vGRF) utilizing consumer-grade sensors, specifically the Apple Watch. This Data Descriptor highlights the importance of integrating accessible technology in biomechanical research, providing valuable insights for both researchers and practitioners in the field.
The dataset comprises recordings from ten healthy adults, aged between 26 and 41 years, who performed five distinct activities: walking, jogging, running, heel drops, and step drops. Each participant donned two Apple Watches—one positioned on the left wrist and the other at the waist—while their movements were simultaneously monitored using a laboratory force plate to establish ground truth data.
Dataset Overview
In total, the dataset consists of 492 validated trials that include time-aligned inertial measurement unit (IMU) recordings at an approximate frequency of 100 Hz, alongside force plate vGRF data sampled at 1000 Hz. Key features of the dataset include:
- Raw and processed time series data.
- Trial-level metadata and quality-control flags.
- Machine-readable data dictionaries for ease of use.
A notable aspect of this dataset is the trial-level matching manifests, which effectively link recordings across different modalities through stable identifiers. Out of the 492 validated trials, 395 are triad-complete, meaning they contain synchronized data from the wrist, waist, and force plate. This completeness allows for robust cross-sensor analyses and reproducible model evaluation, making it a significant resource for researchers in wearable biomechanics.
Quality and Consistency Assessment
The quality of the dataset has been rigorously assessed through a three-phase cross-sensor plausibility and consistency framework. Additionally, a repeatability analysis of peak vGRF yielded an intraclass correlation coefficient ranging from 0.871 to 0.990, indicating high reliability. Systematic checks of force ranges and trial completeness further contribute to the dataset’s integrity.
A Monte Carlo sensitivity analysis also demonstrated that correlation-based validation metrics remained robust despite single-sample timing perturbations at the IMU sampling resolution, ensuring the dataset’s reliability for future research.
Accessibility and Future Research
All data included in this dataset is released under the Creative Commons BY 4.0 license, promoting open access and collaboration within the research community. Additionally, analysis scripts are archived alongside the dataset and are mirrored on GitHub, facilitating easier access for researchers aiming to replicate findings or develop new methodologies in the domain of wearable technology and biomechanics.
This resource stands to support reproducible research in wearable biomechanics, the benchmarking of machine learning models for vGRF estimation, and the investigation of sensor placement effects using widely available consumer wearables. As technology continues to evolve, such datasets are crucial for advancing the field and enhancing our understanding of human movement.