Speaker
Description
This study presents a data-driven framework for analysing human gait as a dynamical system using multimodal sensor data. The approach combines high-resolution plantar pressure measurements—i.e., the spatial distribution of contact pressure between the foot and the ground—obtained from instrumented insoles, with time-synchronised physiological signals and contextual annotations across multiple repeated measurements in real-world conditions.
Data acquisition was performed using wearable pressure sensors capturing spatio-temporal load distribution across 16 channels (8 per foot), enabling reconstruction of step sequences, load transfer patterns, and centre-of-pressure dynamics. Measurements were conducted under varying conditions, including baseline walking, externally influenced contexts (e.g. footwear changes and environmental conditions), and repeated trials per subject, resulting in a dataset with high intra-individual variability.
The analysis pipeline comprises:
1. preprocessing and temporal alignment of sensor streams,
2. extraction of gait features (step timing, pressure integrals, spatial distribution),
3. computation of variability metrics (intra-step and inter-step variance), and
4. evaluation of condition-dependent changes across repeated measurements.
Results indicate that gait patterns exhibit structured variability rather than random noise. Across repeated trials, individuals demonstrate consistent baseline signatures, whilst condition changes induce reproducible shifts in pressure distribution, step dynamics, and variability measures. Both increases and decreases in variability were observed depending on the type of perturbation, suggesting that variability is context-dependent rather than unidirectional.
These findings support the interpretation of gait as a dynamically stable system operating within a bounded region of variability, where deviations reflect adaptive responses to internal and external constraints rather than necessarily pathological states.