Decoding Brain Dynamics in Motor Planning Based on EEG Microstates for Predicting Pedestrian Road-Crossing in Vehicle-to-Everything Architectures

2024-05-22

AI Summary

Autonomous vehicles struggle to respond to sudden pedestrian crossings, especially under occlusion conditions. Method: This study proposes a wearable EEG-based motion-intention prediction method that models the dynamic evolution of EEG microstates. It introduces, for the first time, a coupling of Gaussian Hidden Markov Models (GHMMs) with motor-preparation decoding to establish a cross-subject, low-latency (~1 s) brain-to-vehicle direct interface. Leveraging time-frequency features from the central high-beta band, functional connectivity, and sliding-window EEG data, the approach integrates k-nearest neighbors (k-NN) and dynamic time warping (DTW) classification. Results: The system achieves an AUC of 0.91 and predicts physical crossing actions on average 1.02 seconds in advance. This work validates EEG as a feasible, robust, and timely active intent source for V2X systems, offering a novel paradigm to enhance autonomous vehicle responsiveness to occluded pedestrians.

Technology Category

• Intelligent Robots: State Estimation
• Humans and AI: Brain-Sensing and Analysis
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Pedestrians who cross roads, often emerge from occlusion or abruptly begin crossing from a standstill, frequently leading to unintended collisions with vehicular traffic that result in accidents and interruptions. Existing studies have predominantly relied on external network sensing and observational data to anticipate pedestrian motion. However, these methods are post hoc, reducing the vehicles' ability to respond in a timely manner. This study addresses these gaps by introducing a novel data stream and analytical framework derived from pedestrians' wearable electroencephalogram (EEG) signals to predict motor planning in road crossings. Experiments were conducted where participants were embodied in a visual avatar as pedestrians and interacted with varying traffic volumes, marked crosswalks, and traffic signals. To understand how human cognitive modules flexibly interplay with hemispheric asymmetries in functional specialization, we analyzed time-frequency representation and functional connectivity using collected EEG signals and constructed a Gaussian Hidden Markov Model to decompose EEG sequences into cognitive microstate transitions based on posterior probabilistic reasoning. Subsequently, datasets were constructed using a sliding window approach, and motor readiness was predicted using the K-nearest Neighbors algorithm combined with Dynamic Time Warping. Results showed that high-beta oscillations in the frontocentral cortex achieved an Area Under the Curve of 0.91 with approximately a 1-second anticipatory lead window before physical road crossing movement occurred. These preliminary results signify a transformative shift towards pedestrians proactively signaling their motor intentions to autonomous vehicles within intelligent V2X systems. The proposed framework is also adaptable to various human-robot interactions, enabling seamless collaboration in dynamic mobile environments.

Problem

• Predict pedestrian road-crossing intent using EEG signals
• Analyze brain dynamics for motor planning in traffic scenarios
• Enhance V2X systems with proactive pedestrian intention signaling

Innovation

• EEG microstates decode brain dynamics for prediction
• Gaussian Hidden Markov Model analyzes cognitive transitions
• K-nearest Neighbors with Dynamic Time Warping predicts movement