Perceived sAfety and Comfort zone boundaries to improve user Experience and Safety of autonomous driving and driving assistance systems (PACES)

Research Project, 2025 – 2027

As autonomous driving systems (ADS) and driver assistance systems (DAS) mature, the focus is shifting from crash avoidance to conflict avoidance. It is increasingly important to develop these systems not only to be safe, but to be accepted by drivers, ideally eliciting a "thank you" response when a system performs a conflict or crash avoidance maneuver. Research has shown that driver acceptance is strongly related to perceived safety—the safer a system is perceived to be, the more drivers accept and trust it. Recent research from TU Delft has made substantial progress in quantifying perceived safety in terms of the interplay between the ego vehicle and surrounding vehicles (kinematics from the ego-vehicle driver’s perspective). Study participants (from driving simulator experiments and crowdsourcing) rated their perceived safety, which was then modeled using machine learning and explainable AI methods to establish a relationship between kinematics and perceived risk. The aim of this post-doc project is to further develop the TU Delft method to: a) enhance perceived-safety predictability, b) increase the time resolution of the predictions, c) operationalize comfort zone boundaries (CZB) in terms of perceived safety, d) further develop and assess perceived safety and CZB models using naturalistic driving data, and e) extend the scenario coverage to include vulnerable road users. The intended users of the results are developers of ADS and DAS systems for conflict and crash avoidance, in their quest to improve the systems with respect to both user experience and safety. Increasing perceived safety will not only improve user experience (i.e., acceptance and trust) but also enhance safety system performance by incorporating CZBs into the algorithms for such systems.