Optimal lateral acceleration for different levels of automated driving: a test track study of passenger evaluation for curve negotiation

Lateral acceleration impacts passenger comfort in automated vehicles (AVs). This study investigated the optimal lateral acceleration range for AVs negotiating curves, considering curve radii, engagement in non-driving related activities (NDRAs), and automation level (SAE L1 vs L4). An experiment was conducted on a test track with 28 participants. Participants, seated in the driver's seat, experienced three lateral acceleration levels (2, 3, 4 m/s2) across five different road curves. Evaluations, using a 7-point bipolar scale, were collected for both SAE Level 1 (assisted driving) and Level 4 (highly automated driving), with NDRA engagement considered under L4. Data were analysed using multilevel Bayesian multinomial logistic models. Results suggest that no single universally “optimal” lateral acceleration was identified, as preferences were contextual. Generally, 2 m/s2 was often rated “too slow”, particularly without NDRAs in L4, or in L1 driving. Conversely, 4 m/s2 was frequently perceived as “too fast”. Within the tested range, the level of 3 m/s2 often appeared to be a more balanced option, provided that road geometry, passenger activity, and automation level are accounted for. However, its optimality depended on other factors. Larger curve radii (gentler curves) led to a preference for lower lateral accelerations. NDRA engagement in L4 driving made 2 m/s2 receive fewer “too slow” ratings and amplified the “too fast” perception for higher accelerations. L4 automation, compared to L1, also made 2 m/s2 receive fewer “too slow” ratings. The findings provide detailed guidelines for designing AV lateral control systems, suggesting that lateral acceleration should be dynamically adjusted based on contextual factors to ensure comfort throughout the journey.