A Pilot Study on Falling-Risk Detection Method Based on Postural Perturbation Evoked Potential Features.

Area of Science:

• Neuroscience
• Robotics
• Human-Computer Interaction

Background:

• Human-robot systems face falling risks due to erroneous motor execution.
• Existing fall detection methods rely on computer vision and physical signals, neglecting neural activity.
• Human detection of errors is reflected in central nervous system activity.

Purpose of the Study:

• To propose and validate a novel method for monitoring erroneous motion events using electroencephalogram (EEG) features.
• To assess the feasibility of using neural responses for detecting dangerous fall events in real-time.
• To investigate the generalization ability of a fall-risk detection model based on neural activity.

Main Methods:

• 15 subjects received unpredictable postural perturbations while standing.
• EEG signals were analyzed, focusing on features evoked by postural perturbations.
• The xDAWN algorithm reduced EEG signal dimensionality, and Bayesian Linear Discriminant Analysis (BLDA) trained a classifier.
• Detection rate and latency were evaluated against a 90% detection rate threshold.

Main Results:

• EEG analysis revealed a significant negative peak at 62 ms post-perturbation (-14.75 ± 5.99 μV).
• The developed classifier achieved a 98.67% detection rate for falling-risk onset with a 334ms latency.
• The model demonstrated good generalization, achieving a 94.2% detection rate for unlearned atypical perturbation events.

Conclusions:

• Neural activity, specifically EEG features, can be effectively utilized for detecting falling risk in human-robot interactions.
• The proposed method offers a novel, brain-based approach to fall detection with high accuracy and a rapid response time.
• This research paves the way for more sophisticated and responsive human-robot safety systems.