Situation assessment and semantic maneuver planning under consideration of uncertainties for cooperative vehicles in heterogeneous traffic scenarios

后续申请“异构交通场景中考虑不确定性的协同车辆态势评估和语义机动规划”的目标是研究异构场景中基于真实的数据的隐式协同态势评估和隐式协同决策的机器学习方法。为了提高自动驾驶汽车的预见能力，从而主动避免紧急情况的发生，需要根据实际情况进行必要的概括。虽然自动驾驶汽车的能力在不断发展，但它们暂时还缺乏一个将其与人类区分开来的重要组成部分--（隐式）合作能力。与今天的自动驾驶汽车不同，人类驾驶员在他们的决策中包括其他驾驶员的（微妙的）行动和意图，因此即使没有明确的沟通，也能够要求或提供合作。虽然许多研究项目已经解决了合作驱动，特别是在最近几年，重点是明确的合作。由于在可预见的未来，所有车辆都不会有必要的技术解决方案来实现车辆之间以及与基础设施的通信，算法也不会标准化到统一考虑所传达的环境信息和行为决策的程度，因此，即使没有通信，自动驾驶车辆也应该能够与其他车辆合作。正在研究不需要显式通信并且因此可以应用于真实的世界中的交通状况的概念和方法。此外，该问题的建模允许将情况评估和决策扩展到骑自行车的人以及行人的异构场景，这是专门研究的。由于自动驾驶，特别是隐式合作是高度复杂的，它是不切实际的完全建模相应的理解或行为的专家。相反，重点是机器学习方法，以解决评估情况和做出决策的挑战。当前的机器学习方法提供了在高维状态空间上泛化的可能性，从而降低了在不同交通情况下理解和决策的复杂性。尽管安全学习是一个主要挑战，这些方法可用于通过将它们用作算法或初始解来加速模型驱动方法的收敛。

The aim of the subsequent application “Situation assessment and semantic maneuver planning under consideration of uncertainties for cooperative vehicles in heterogeneous traffic scenarios” is the research of machine learning methods using real data for implicit cooperative situation assessment and implicit cooperative decision-making in heterogeneous scenarios. This should be done by establishing the necessary generalization with reference to reality in order to improve the anticipatory capabilities of automated vehicles and thus proactively avoid critical situations.While the capabilities of automated vehicles are constantly evolving, they still lack an essential component that will distinguish them from humans for the time being - the ability of (implicit) cooperation. Unlike today's automated vehicles, human drivers include the (subtle) actions and intentions of other drivers in their decisions, and are thus able to demand or offer cooperation even without explicit communication. Although many research projects have addressed cooperative driving, especially in recent years, the focus is on explicit cooperation. Since neither all vehicles will have the necessary technical solution in the foreseeable future to enable communication between vehicles and with the infrastructure, nor will algorithms be standardized to such an extent that communicated environmental information and behavioral decisions will be considered uniformly, automated vehicles should be able to cooperate with other vehicles even without communication.For this reason, concepts and methods are being researched that do not require explicit communication and can therefore be applied to traffic situations in the real world. Furthermore, the modeling of the problem allows an extension of the situation assessment and decision-making to heterogeneous scenarios with cyclists as well as pedestrians, which are specifically being researched.Since automated driving and especially implicit cooperation are highly complex, it is not practicable to completely model the corresponding understanding or behavior by experts. Instead, the focus is on machine learning methods to solve the challenges of assessing the situation and making decisions.Current machine learning approaches offer the possibility to generalize over a high-dimensional state space and thus reduce the complexity of understanding and decision making in different traffic situations.Although safe learning is a major challenge, these methods can be used to accelerate the convergence of model-driven approaches by using them as heuristics or as an initial solution.