This study formulates a latent class-based driving behaviour framework for modelling vehicles’ two-dimensional (2D) movements while considering drivers’ strategic intents and multi-vehicle anticipation (MVA) in heterogeneous, disorderly (HD) traffic conditions. Specifically, five extensions are proposed to a typical stimulus–response based driving behaviour framework. First, the subject vehicle’s 2D movements are represented as a combination of the angular direction of movement with respect to the longitudinal axis and the magnitude of acceleration or deceleration along the angle. Second, a latent class framework is used to recognise drivers’ strategic intents (latent to the analyst) in two aspects: (a) the intent to accelerate, decelerate, or maintain a constant speed, and (b) the intent to steer to the left of, right of, or straight along the longitudinal axis. It is hypothesised that these strategic intents precede tactical decisions, such as how much to accelerate or decelerate and which specific angular direction to move along. Third, the MVA effect is accommodated to recognise that drivers consider stimuli from multiple vehicles in their vicinity. Fourth, a multi-stimuli model of acceleration (deceleration) is formulated, assuming that drivers choose an angle of movement that allows them to move with the highest (lowest) possible longitudinal acceleration (deceleration). Fifth, drivers’ execution errors are recognised as the difference between their planned acceleration and executed acceleration. The proposed framework is applied for an analysis of motorised two-wheeler driver behaviour using a vehicular trajectory dataset from India. The empirical results highlight the importance of incorporating MVA and considering driver’s intents while modelling 2D movements of vehicles in HD traffic conditions. Further, the microscopic traffic environment variables are found to have a stronger influence on drivers’ higher-level, strategic intents than on their lower-level, tactical decisions.

中文翻译：

---

基于二维、多车辆预期和多刺激的潜在类框架，用于对异构、无序交通条件下的驾驶员行为进行建模

本研究制定了一个基于潜在类别的驾驶行为框架，用于对车辆的二维 (2D) 运动进行建模，同时考虑异构、无序 (HD) 交通条件下驾驶员的战略意图和多车辆预期 (MVA)。具体来说，对典型的基于刺激响应的驾驶行为框架提出了五种扩展。首先，主体车辆的 2D 运动表示为相对于纵轴的运动角度方向和沿该角度的加速度或减速度大小的组合。其次，潜在类别框架用于从两个方面识别驾驶员的战略意图（对分析师来说是潜在的）：（a）加速、减速或保持恒定速度的意图，以及（b）转向的意图纵向轴线的左侧、右侧或沿纵向轴线的直线。假设这些战略意图先于战术决策，例如加速或减速多少以及沿着哪个特定角度方向移动。第三，适应 MVA 效应以识别驾驶员考虑来自附近多辆车的刺激。第四，假设驾驶员选择允许他们以尽可能高（低）的纵向加速度（减速度）移动的运动角度，则制定了加速度（减速度）的多刺激模型。第五，驾驶员的执行误差被认为是他们的计划加速度和执行加速度之间的差异。所提出的框架适用于使用印度的车辆轨迹数据集分析机动两轮车驾驶员的行为。实证结果强调了在高清交通条件下对车辆的 2D 运动进行建模时纳入 MVA 并考虑驾驶员意图的重要性。此外，微观交通环境变量对驾驶员的高层战略意图的影响比对低层战术决策的影响更大。