Model Predictive Control for Integrated Motion Planning and Control of Automated Vehicles

本项目的主要目标是改进自动车辆的模型预测控制技术。在一些研究项目中，预测控制已被认为是解决轨迹跟踪问题的一种很有前途的方法。它的主要优点是对带约束的非线性多变量控制问题的整体表述，具有通用性和直观性。然而，由于几个众所周知的原因，MPC尚未进入工业生产。最重要的是，MPC在计算上相当昂贵，并且需要很高的实现技能。因此，本项目旨在缓解这些弊端的同时，进一步强化MPC的优势。特别是，重点将放在基于情景的MPC(SCMPC)方法上，申请者在过去几年里一直在加紧研究这一方法。这项工作将继续下去，以加强和扩大SCMPC的基础理论。此外，MPC的范围应扩大到包括轨迹跟踪控制和运动规划两项基本任务。与其他方法相比，这将简化底层软件体系结构并减少所需的接口。引入约束的预测控制的特殊强度可用于显式地推导闭环系统的安全保证。为了避免自动车辆行为中不必要的保守，必须对周围交通的未来行为进行解释和预测。为此，SCMPC框架使用场景来预测其他交通参与者的行为。主要思想是，可以使用一组场景(或粒子)来表示其他代理未来可能的行为及其固有的不确定性。这种方法在计算上高效、直观、易于实现，因为它可以直接从经验驾驶数据中提取。在本项目的第二部分中，将开发一个用户友好的工具箱，以便于在广泛的硬件系统上实施MPC。这个工具箱将显著降低MPC的计算负担，因为它建立在高效的最先进算法的基础上，这些算法是专门为集成车辆运动规划和控制任务量身定做的。该工具箱包括可通过图形用户界面(GUI)操作的代码生成器。因此，该项目抽象了实现过程中的大部分困难，可以为控制器生成独立的C代码，可以在大多数嵌入式平台上进行集成和测试，总之，该项目将扩大MPC在自动驾驶方面的技术水平，并将显著简化其工业应用。

The main goal of this project is to improve the technology of Model Predictive Control (MPC) for automated vehicles. MPC has been identified as a promising approach for the trajectory tracking problem in several research projects. Its key advantage is the holistic formulation of nonlinear multi-variable control problems with constraints, which is versatile and intuitive. However, MPC has not found access into industrial production yet, for several well-known reasons. Most importantly, MPC is computationally rather expensive and it requires a high degree of implementation skills. Therefore this project aims at mitigating these drawbacks, while further strengthening the advantages of MPC. In particular, the focus will be on the approach of Scenario-based MPC (SCMPC), on which the applicant has worked intensively over the past years. This work shall be continued in order to strengthen and expand the fundamental theory of SCMPC. Moreover, the scope of MPC shall be extended to comprise the two basic tasks of trajectory tracking control and motion planning. This leads to a simplification of the underlying software architecture and a reduction of the required interfaces compared to other approaches. The particular strength of MPC of incorporating constraints can be used to explicitly derive safety guarantees for the closed-loop system. To avoid unnecessary conservatism in the behavior of the automated vehicle, the future behavior of the surrounding traffic has to be interpreted and anticipated. To this end, the framework of SCMPC uses scenarios for the prediction of the behavior of other traffic participants. The main idea is that a bundle of scenarios (or particles) can be used to express the likely future behavior of other agents as well as its inherent uncertainty. This approach is computationally efficient, intuitive, and easy to implement, because it can draw directly from empirical driving data. Explicit safety guarantees can be derived by the theory on scenario-based optimization.In the second part of this project, a user-friendly toolbox will be developed that facilitates the implementation of MPC on a wide range of hardware systems. This toolbox will significantly reduce the computational burden of MPC, as it builds on efficient state-of-the-art algorithms that are specifically tailored for the task of integrated vehicle motion planning and control. The toolbox includes a code generator that can be operated via a graphical user interface (GUI). Thus it abstracts most of the difficulties of the implementation process and standalone C code can be produced for the controller, which can be integrated and tested on most embedded platforms.In summary, this project will expand the current state of technology of MPC with respect to automated driving, and it will significantly simplify its industrial application.