A cognitive digital twin approach to improving driver compliance and accident prevention

Advanced Driver Assistance Systems (ADAS) are crucial for enhancing driving safety by alerting drivers to unrecognized risks. However, traditional ADAS often fail to account for individual decision-making processes, including drivers' perceptions of the environment and personal driving styles, which can lead to non-compliance with the provided assistance. This paper introduces a novel Cognitive-Digital-Twin-based Driving Assistance System (CDAS), leveraging a personalized driving decision model that dynamically updates based on the driver's control and observation actions. By incorporating these individual behaviors, CDAS can tailor its assistance options to predict and adapt to the driver's responses across various scenarios, ensuring both the necessity and safety of its interventions. Through two comprehensive experimental validations, we demonstrate that the cognitive digital twin (CDT) closely aligns with actual driver observation behaviors. By incorporating additional driver observation actions - an input not readily leveraged by data-driven methods without large annotated datasets - the CDT also achieves superior lane-changing predictions compared to deep learning classifiers relying solely on environmental states. Furthermore, CDAS significantly outperforms traditional ADAS in terms of risk reduction and user acceptance, showcasing its potential to enhance driving safety and adaptability effectively. These findings suggest that CDAS represents a substantial advancement towards more personalized and effective driving assistance.