Neurocognitive and Experiential Evaluation of Autonomous Driving Strategies Using EEG and a Four-Dimensional Human Experience Model

Abstract

As autonomous driving technologies continue to evolve, understanding passengers' cognitive and experiential responses has become increasingly important for human-centered system design. However, the ways in which different autonomous driving decision-making paradigms influence passenger experience and underlying neurocognitive processes remain insufficiently explored. This study proposes an EEG-based evaluation framework grounded in a Four-Dimensional Human Autonomous Driving Experience Model (4D-HADEM), integrating comfort, tension, perceived control, and trust. Using the CARLA simulation platform, we conducted a within-subject experiment with 28 participants, in which three representative autonomous driving strategies—modular pipeline (MP), imitation learning (IL), and reinforcement learning (RL)—were adopted to model different levels of autonomous driving performance across three urban driving scenarios. Multimodal data, including EEG signals, vehicle-level metrics, and subjective evaluations, were analyzed with attention to their statistical hierarchy. Results showed that driving strategy and scenario significantly affected comfort, perceived control, and trust, with MP generally yielding the most favorable subjective profile. EEG analysis based on theta, alpha, and beta bands revealed significant model effects in occipital alpha and beta power, a scenario effect in parietal alpha power, and a Model × Scenario interaction in occipital beta power after FDR correction. Vehicle behavior was examined condition level, and its associations with subjective ratings and EEG measures were therefore evaluated at the same level. Repeated-measures correlations further showed positive associations between alpha activity and comfort, trust, and perceived control, although these associations did not remain FDR-significant after controlling for model and scenario. These findings demonstrate the value of multimodal neuroergonomic evaluation and provide a basis for future adaptive autonomous driving systems that account for passengers’ cognitive states, trust, and riding experience.