Real-time monitoring of driver cognitive states: linking driving behavior to neural oscillations

• EEG detects proactive/reactive neural control linked to driver steering demands. • High variation steering triggers alpha/theta suppression and beta rebound. • Increased steering deviations correlate with alpha and beta suppression. • EEG–eye–vehicle integration offers early detection of error-prone driving states. Human error, from attentional lapses or fatigue, underlies most road crashes. Current driver-monitoring systems primarily track behavioral and ocular signals and therefore fail to capture the cognitive dynamics that precede breakdowns. This study identifies neural oscillatory markers of steering-control demands and their links to gaze and vehicle inputs for real-time monitoring. Thirty-two drivers completed a simulated highway car-following task under three visibility scenarios (sunny, rainy, foggy) and two workload levels (low vs. high overtaking demand). Continuous EEG, eye movements, and vehicle-control signals were recorded synchronously. A principal-component-derived variation metric on steering, lane position, and pedal signals segmented epochs into low- and high-variation steering events. We quantified frontal theta, posterior alpha, and sensorimotor beta power around each event using time–frequency analyses; eye and vehicle indices were extracted in the same windows. Low-variation events showed sustained pre-event posterior-alpha, frontal-theta and sensorimotor-beta activity, whereas high-variation segments showed rhythm suppression followed by a post-event sensorimotor-beta rebound. High-variation events showed stronger control corrections (steering deflection ≈0.02° vs. ≈0.002°; brake pulse ≈0.30 vs. ≈0.10; speed drop from ≈106 to ≈100 km/h within 1 s) and greater arousal (pupil dilation ≈ 0.1 mm). Neural-behavioral coupling was graded: larger steering deviations and braking correlated with alpha and beta suppression (ρ = −0.32 to −0.42, FDR-adjusted p < 0.05). High-variation events also increased saccade amplitude and count without altering fixation duration, indicating broadened visual sampling. These dynamics support proactive versus reactive dual-mode control and suggest that multimodal EEG-ocular-vehicle monitoring can detect error-prone driver states early.