This study examines the development of drivers' general mental models during their first real-world experience with the SAE Level 3 conditionally automated driving system (CADS) Drive Pilot. While previous research has primarily investigated mental model formation in simulators or on test tracks, little is known about how accuracy and completeness evolve during initial use in naturalistic traffic. Twenty-nine participants without prior CADS experience completed a within-subject on-road study with three measurement points: before receiving any information about the CADS (t 1 ), after a short instructional video (t 2 ), and after a real-world drive on a German motorway (t 3 ). Mental models were assessed with a system-specific self-report questionnaire designed to evaluate both accuracy and completeness. Qualitative and statistical analyses showed high initial accuracy for core functions, alongside considerable misconceptions and knowledge gaps regarding limitations and operational aspects. The instructional video improved both accuracy and completeness, including for some limitations not explicitly covered. Real-world driving further increased accuracy across categories. However, completeness declined, particularly for limitations not encountered during the drive. Statistical analyses confirmed significant improvements in accuracy from t 1 to t 2 , t 1 to t 3 and t 2 to t 3 . Findings suggest that short, targeted instructions combined with immediate real-world exposure can effectively enhance the accuracy of drivers' mental models. However, knowledge about seldom-encountered limitations decays rapidly without reinforcement, highlighting the need for specific instruction and in-vehicle systems that sustain awareness of rare but safety-critical constraints over time. • Real-world study on development of drivers' mental model of conditionally ADS. • Short instruction video and first drive significantly enhanced mental model accuracy. • Mental model completeness decreased, especially for rarely encountered system limits. • Indirect transfer improved understanding of features not directly experienced. • Findings highlight need for short trainings and HMI that reinforce rare constraints.
Schwindt-Drews et al. (Wed,) studied this question.