Complexity in cooperative traffic situations: Discrete-event modeling in automated vehicle guidance

研究的问题。该项目涉及的主题是情况预测和系统人机工程学。在许多交通情况下，尽管存在有效的交通规则，但由于缺乏合作行为，会导致交通崩溃。这种死锁情况经常发生在城市的X型和t型十字路口或同等狭窄的通道中。此外，我们之前的工作表明，当交通状况的解决方式与预期不同时，合作需求也可能自发产生。进一步，假设交通状况的复杂性对人类的合作行为有重要影响。我们的目标是更好地理解人类在这种情况下的行为：哪些环境信息与评估合作环境有关？如何认识合作伙伴的行为意向？在此基础上，推导出一种态势预测算法，使自动驾驶汽车也能在与人类驾驶员的合作下克服死锁情况。研究方法和结果。为了实现这些目标，将进行实证研究：除了在“Testfeld Autonomes Fahren baden - w<s:1> rttemberg”进行现场实验研究和观察性交通研究外，还计划在驾驶模拟器中进行进一步的实验。由于我们假设人类的合作行为高度依赖于交通复杂性，因此应该提出一种允许自动评估的合适度量。在此基础上，建立事件离散的合作情景描述的时间依赖随机模型，以预测合作伙伴可能的意图，并为车内乘客和其他道路使用者都接受的最优行为决策做准备。最后，算法将在实时框架内实现，并在测试车辆上进行演示。作为观测交通研究的结果，预计将产生一个庞大的数据库，并将提供给其他研究小组。

Research question. The project addresses the topics situation prediction and system ergonomics. In many traffic situations a breakdown will be observed when cooperation behavior is missing, despite of valid traffic rules. Such deadlock situations occur in urban X- and T-intersections or in coequal narrow passages regularly. Besides, our previous work has shown that a need of cooperation may also arise spontaneously when a traffic situation is solved differently than it is supposed to be. Furthermore, it is to be assumed that the complexity of a traffic situation has an important impact on the human cooperation behavior. Our objective is to gain a better understanding of how humans behave in such situations: Which environmental information is relevant for assessing a cooperative setting? How do we recognize the behavioral intention of cooperation partners? On the basis of these findings, an algorithm for situation prediction shall be derived, so that automatic vehicles are also able to overcome deadlock situations in cooperation with human drivers. Research methods and results. In order to achieve these goals, empirical studies are to be carried out: Besides a field experimental study and an observational traffic study within „Testfeld Autonomes Fahren Baden-Württemberg“ further experiments in a driving simulator are planned. As we assume that human cooperation behavior depends highly on traffic complexity, a suitable measure shall be proposed that allows for an automatic assessment. On this basis time-dependent stochastic models for the event-discrete description of cooperation situations shall be developed, in order to predict the likely intentions of cooperation partners and to prepare for an optimal behavioral decision, accepted by both the passengers in the car and other road users. Finally, the algorithm shall be implemented within a real-time framework and it will also be demonstrated in a testing vehicle. As an outcome of the observational traffic study a huge database is to be expected, which shall be made available to other research groups.