Ground Perturbation Detection via Lower-Limb Kinematic States During Locomotion

TL;DR

Falls during daily ambulation threaten safety in older adults; WBAM-based perturbation detection is not well-suited to real-time exoskeleton control due to computational delays and the need for normalization. The authors introduce a data-driven detector that tracks 16 lower-limb kinematic states in a local coordinate system and flags perturbations with a single threshold $\varphi$ based on deviations from steady-state gait. On an open-source dataset, the method achieves $87.65\%$ accuracy with a $28.12\%$ gait-cycle delay; in a pilot study with five subjects, accuracy reaches $98.8\%$ with about a $23\%$ delay, outperforming the WBAM baseline in detection accuracy. The work demonstrates a subject-independent, computationally efficient approach with strong potential to enhance real-time control of balance-supporting exoskeletons, particularly for older adults.

Abstract

Falls during daily ambulation activities are a leading cause of injury in older adults due to delayed physiological responses to disturbances of balance. Lower-limb exoskeletons have the potential to mitigate fall incidents by detecting and reacting to perturbations before the user. Although commonly used, the standard metric for perturbation detection, whole-body angular momentum, is poorly suited for exoskeleton applications due to computational delays and additional tunings. To address this, we developed a novel ground perturbation detector using lower-limb kinematic states during locomotion. To identify perturbations, we tracked deviations in the kinematic states from their nominal steady-state trajectories. Using a data-driven approach, we further optimized our detector with an open-source ground perturbation biomechanics dataset. A pilot experimental validation with five able-bodied subjects demonstrated that our model distinguished perturbed from unperturbed gait cycles with 98.8% accuracy and only a delay of 23.1% within the gait cycle, outperforming the benchmark by 47.7% in detection accuracy. The results of our study offer exciting promise for our detector and its potential utility to enhance the controllability of robotic assistive exoskeletons.

Figures (6)

• Figure 1: Kinematic state-based perturbation detection. The relative distance between the body's COM and each foot was calculated to track the kinematic state (right). Eight relative position states and eight velocity states were tracked. A local coordinate system was established at the estimated midpoint between the feet in the horizontal plane. The COM position was then projected into this coordinate system, with a vector (red dashed lines) tracking the relative distance between the COM and the feet over time (left).
• Figure 2: (A) Principal Component Analysis for a single representative trial of an (A) anteroposterior and (B) mediolateral perturbation. For both cases, perturbations with a magnitude of 5 cm occurred during the double stance phase. (C) Visual representation of state variance calculation (Eqs. \ref{['eq:eq1']} and 2). $\alpha$ is the Euclidean distance between the current state value and the edge of 2 standard deviations ($\sigma$) from the mean.
• Figure 3: (A) Experimental setup. (B) Split-belt treadmill protocol for trip- and slip-type perturbations. (C) Perturbation timing: early stance (0-10% and 90-100% of the gait cycle), mid stance (10-30% of the gait cycle), late stance (30-50% of the gait cycle), and swing phase (50-90% of the gait cycle) of the right leg. Error bars indicate $\pm$1 standard deviation. (D) Visual illustration of tracking of positional states. Blue dots represent expected positional states and red dots show actual states values during a perturbation. Black arrows indicate a distance vector between expected and actual states.
• Figure 4: Parameter sweep result to determine an optimal perturbation detection threshold, $\varphi$. A threshold value of 0.125 (black dashed lines) was found to be the optimal value.
• Figure 5: (A) A representative trial of our model performance in real-time. The red line represents the perturbation onset and the green line represents the time when our model detects the perturbation (detection threshold shown with the dashed grey line). (B) A confusion matrix summarizing our model performance. Positive and negative labels indicate perturbed and unperturbed gait cycles, respectively.