GROKO-Plan: GRaph-based, Optimal and COoperative Trajectory planning for Interacting Automobiles

该项目旨在开发一种基于图的交互车辆最优合作轨迹规划方法。本文的研究包括分布式协同轨迹规划方法的理论分析和实际实现，并将在不同的测试平台上对协同交通交叉口场景进行验证，分布式协同轨迹规划的主要挑战是规划轨迹的可靠性(I)、车辆上优化器的实时一致性(II)和车辆之间可实现的通信努力(III)。由于无冲突轨道是强制性的，因此出现了第一个挑战。第二个挑战是高维非凸优化问题，因为它们发生在必须考虑许多道路使用者的情况下。第三个挑战来自于合作的蓬勃发展，即要解决一辆车上的协同优化问题，必须交换其他道路使用者的部分优化问题。为了应对这些挑战，本项目提出使用由机动和Trim基元组成的简化轨迹规划库来进行协同轨迹规划。然后，通过在简化的协作图上进行分布式模型预测控制(DMPC)来执行轨迹规划。稳健的DMPC方法考虑了感知和预测的不确定性。由此，可以生成无冲突轨迹(I)。总而言之，规划方法提供的轨迹是动作和修剪基元的序列。这大大减少了搜索空间，并实现了实时优化(Ii)。与现有的基于图的路径规划方法不同，新的单车轨迹规划方法是协同工作的，并且考虑了相邻车辆的目标和约束。在沟通之前减少图表对我们的方法至关重要。它简化了优化问题的复杂性(II)，并使通信努力变得可接受(III)。本项目内的新算法的模拟验证基于SPP在第一资金阶段开发的模拟平台。特别是，将使用开发的协作感知和预测算法以及交通场景存档。反过来，该仿真平台将根据该项目的结果进行扩展。我们将利用已经并将在战略规划进程中制定的评估指标和基准系统，例如CommonRoad。

This project aims at developing a graph-based planning method for optimal cooperative trajectories for interacting vehicles. Our research includes the theoretical analysis as well as the practical implementation of a distributed cooperative trajectory planning method, which will be validated for a cooperative traffic intersection scenario on different test platforms.The major challenges of distributed cooperative trajectory planning are dependability of planned trajectories (I), real-time compliance of the optimizer on board of the vehicles (II), and a realizable communication effort between vehicles (III). The first challenge arises since conflict-free trajectories are mandatory. The second challenge is due to high-dimensional non-convex optimization problems, as they occur when many road users have to be taken into account. The third challenge is caused by the thrive for cooperation, i.e. to solve the cooperative optimization problem on one vehicle, parts of the optimization problems of other road users have to be exchanged.In order to meet these challenges, this project proposes to use reduced trajectory planning libraries, comprised of maneuvers and trim primitives, for cooperative trajectory planning. Trajectory planning is then performed by distributed model predictive control (DMPC) working on reduced cooperative graphs. Uncertainty in perception and prediction is considered in robust DMPC methods. Thereby, conflict free trajectories can be generated (I). All together, the planning method provides trajectories which are sequences of maneuvers and trim primitives. This considerably reduces the search space and enables real-time optimization (II). In contrast to existing graph-based approaches, the novel trajectory planning method on an individual vehicle works cooperatively und takes into account objectives and restrictions of neighboring vehicles. Reducing the graphs before communicating them is crucial to our approach. It simplifies the complexity of the optimization problems (II) and the communication effort becomes acceptable (III).The simulative validation of the novel algorithms within this project is based on the simulation platform which has been developed within the SPP during the first funding phase. In particular, developed algorithms for cooperative perception and prediction as well as the archive of traffic scenarios will be used. In turn, the simulation platform will be extended by the results of this project. We will make use of evaluation metrics and benchmark systems, e.g. CommonRoad, which have been and will be developed within the course of the SPP.