The driving scenario and information load jointly influence the cognitive efficiency of augmented reality head-up display (AR-HUD) interfaces. However, the moderating role of drivers’ working memory capacity (WMC) remains unclear. To investigate this mechanism, a simulated driving experiment with a mixed design was conducted in a low-immersivity desktop virtual reality (VR) environment. First, 40 volunteers were screened using an automated operation span task, yielding 16 high- and low-WMC participants. They then drove under three scenarios (urban intersection, expressway, construction zone) and six levels of AR-HUD visual information load. Generalized linear models were applied to the reaction time, fixation duration, and pupil diameter. The results revealed a significant three-way interaction among WMC, scenario, and information load. High-WMC drivers maintained faster responses and lower subjective loads up to Levels 4–6, adopting a deep processing strategy; low-WMC drivers already showed cognitive overload at Level 4 and above, requiring an optimal load range of Level 2–3. The construction zone induced the steepest increase in cognitive load, whereas the expressway markedly reduced sensitivity to additional visual information. Therefore, the optimal AR-HUD information load must be adapted to drivers’ WMC: high-WMC drivers can safely handle Levels 4–6 in low- or medium-demand scenarios, whereas low-WMC drivers require a minimalist presentation of Levels 2–3 in high-demand situations. This study provides quantitative, empirically grounded guidelines for designing cognitively adaptive AR-HUD interfaces.

As augmented reality head-up display (AR-HUD) becomes increasingly integrated into intelligent vehicles, inappropriate interface designs may increase drivers’ cognitive workload and delay hazard responses. This study investigates how cognitive style, driving task type, and AR-HUD navigation design jointly influence drivers’ behavioral performance and visual attention. A total of 55 participants were recruited and screened using the Group Embedded Figures Test, with 38 drivers finally selected for a 2 × 4 × 2 driving-simulation experiment comparing world-fixed (WF) and screen-fixed (SF) interfaces across goal-directed and stimulus-driven tasks. Reaction times and eye-tracking indicators were analyzed using generalized linear models. Results show that stimulus-driven tasks significantly increased reaction times, with rear-vehicle scenarios producing the longest responses (mean = 1.420). During lane-change tasks, WF displays significantly reduced fixation duration (p < 0.001) and fixation counts (p < 0.001), whereas SF displays improved attentional efficiency during pedestrian-warning tasks. In addition, field-dependent drivers exhibited significantly larger pupil diameters, indicating higher cognitive workload. These findings provide sensor-based evidence for AR-HUD systems that dynamically optimize interface presentation according to task context and workload conditions.