What Makes a Driving Situation Complex? An Additive Conjoint Approach to Visual Complexity in Vehicle Teleoperation

Vehicle teleoperation requires operators to perceive, integrate, and evaluate multiple sources of visual information under dynamic and potentially safety-critical traffic conditions. Although previous research has identified several characteristics that may contribute to visual complexity, it remains unclear how strongly individual scene characteristics contribute towards overall perceived complexity when they occur in combination. The present study applied a choice-based conjoint approach with an additive model to estimate the relative contribution of selected traffic-scene characteristics to perceived complexity in vehicle teleoperation. We systematically varied five factors at three levels (low, medium, high): the number of relevant traffic participants, traffic density, recognizability of traffic rules, visibility conditions, and driving situation type. Because the full factorial design would comprise 243 combinations, an orthogonal fractional-factorial L27 design was used to select 27 scenarios. These were presented as real-world driving videos recorded with a three-camera setup. We also included two training scenarios. The study included 28 complete, analyzable datasets (M_age = 24.75, SD_age = 2.52; 12 female, 16 male). The participants completed 68 pairwise comparisons across four blocks and judged which of two videos appeared more complex. They did not rate the underlying attributes separately. Subjective factor importance, decision strategies, perceived difficulty of comparison, and mental demand were assessed additionally. We modeled pairwise judgments using a Bradley-Terry model to derive relative complexity scores for the videos. These scores were then decomposed into effect-coded part-worth utilities, and utility ranges were used to derive the relative importance of the five factors. Initial results suggest that type of situation and relevant traffic participants contributed most strongly to perceived complexity (relative importance: 38% and 28.8%). Visibility contributed 24%, whereas recognizeability of traffic rules (7.6%) and traffic density (1.6%) contributed comparatively little. The subjective rankings showed a less differentiated importance of the manipulated factors, with mean values between 4 and 6 on a scale from 1 to 7. The results may be influenced by the selected scenario representatives and how the factors were operationalized. Still, the study approach offers a systematic way to represent scene-induced complexity beyond isolated traffic characteristics. It may support the operationalization of objective task demands due to scene characteristics for operators and add to a more differentiated understanding of scene-induced complexity. Subsequent studies should use larger samples to stabilize the statistical analysis, refine the factor operationalization, and test generalizability across additional driving scenarios.