Towards situation-aware driving style adaptation

With advancing technological capabilities, the focus of future driver assistance systems and autonomous vehicles is shifting from mere feasibility to implementing acceptable and comfortable driving characteristics for future driver assistance systems and autonomous vehicles. This dissertation explores the heterogeneity of human driving behavior, particularly emphasizing lateral driving behavior and individual driving style preferences on rural roads, fostering research toward more personalized driver assistance systems and autonomous vehicles. A controlled, real-world vehicle study (N = 62) was conducted to identify and define key indicators of driving behavior, analyze sociodemographic dependencies and assess the alignment between subjective self-assessments and objective data. Correlation analyses revealed modest but significant associations between participants’ self-reported driving styles and objective acceleration and jerk statistics. Notably, an in-depth analysis of lateral driving behavior highlighted substantial inter-driver variability. Despite growing interest in adaptive driving styles, comprehensive research on the influence of contextual factors, especially adverse weather conditions, and oncoming traffic, on automated driving functions remains limited. To address this gap, a driving simulator study (N = 42) was carried out to examine lateral driving style preferences for autonomous vehicles under varying weather and traffic conditions. Statistical analyses demonstrated a prevalent preference for passive driving styles, with environmental factors significantly affecting perceived comfort during autonomous rides. To overcome the limitations of existing driving style models in integrating environmental information, this work introduces a situation-aware driving style adaptation method that combines deep learning and statistical approaches. By employing visual feature encoders pretrained on fleet data, the proposed method captures a structured representation of the driving environment and learns a mapping from situational context to driving behavior, enabling adaptation to individual driving style preferences. The developed models significantly outperform baseline approaches and form coherent situation clusters, enhancing the adaptability of automated driving systems. These findings contribute to the development of more intuitive and human-centered automation, facilitating the integration of personalized driving styles into future vehicles while considering real-world environmental influences.

Author dc:creator

• Walser-Haselberger, Johann

Advisor dc:contributor.advisor

• Müller, Steffen

Licence dc:rights.uri

• https://creativecommons.org/licenses/by-sa/4.0/

Language dc:language.iso

en

Identifier URI

https://doi.org/10.14279/depositonce-24767

OAI identifier oai:identifier

oai:depositonce.tu-berlin.de:11303/25940