基于车辆社会行为的无人车混合模型预测控制研究

The social behavior of vehicles refers to the interactions among a swarm of vehicles in a traffic network, and it is vital to safe driving and smooth traffic. In the research of autonomous vehicles, not only does vehicle dynamics need to be considered, but also the interactive behavior among surrounding vehicles should be analyzed. This research studies the hybrid model predictive control of autonomous vehicles based on their social behavior. Firstly, the Markov chain model combined with a Bayesian network model is applied to recognize surrounding vehicles' social behavior. Secondly, a hybrid model is established based on the vehicles' social behavior, and its control law is designed by the theory of hybrid model predictive control. Finally, the real-time optimization mechanism is developed for the control law and its embedded controller is implemented for autonomous vehicles. Through the in-depth research, the hybrid model predictive control is achieved for autonomous vehicles based on their social behavior and the embedded techniques for the control method is further improved.

车辆社会行为是车辆在交通网络中表现出的群体属性，对行车安全及交通顺畅至关重要。无人车驾驶研究不仅需要考虑车辆自身的动力学特性，而且需要分析周边社会车辆与之产生的一系列交互行为。本课题研究基于车辆社会行为的无人车混合模型预测控制。首先，提出基于贝叶斯网络模型的马尔科夫链预测方法，识别周围车辆社会行为；其次，建立基于车辆社会行为的无人车混杂模型，设计混合模型预测控制律；最后，提出控制算法实时性的优化机制，实现无人车嵌入式控制器。通过本课题的深入研究，实现融合车辆社会行为的无人车混合模型预测控制方法，完善控制方法的嵌入式实现技术。

本项目研究基于车辆社会行为的无人车决策与控制的关键问题，旨在提高无人车的智能化水平。首先，本项目提出了基于三维激光雷达的道路边界检测方法和基于多特征地图的无人车定位方法，从而实时识别行驶道路环境并实现车辆的高精度定位；其次，提出了利用时间自动机对车辆跟车社会行为进行学习和建模的方法，实现了跟车社会行为的准确识别，并在此基础上设计了跟车行为生成器，提高了无人车智能化跟车驾驶水平；再次，提出了利用一类特殊的贝叶斯网络——隐马尔科夫模型对变道行为进行学习和建模的方法，提高了无人车对变道行为的预测和识别能力；然后，提出了基于模型预测的无人车动态路径规划和控制算法，实现了多障碍物环境下的无人车路径规划，提高了驾驶性能；最后，构建了无人车硬件在环测试平台，提高了开发测试的效率和安全性，实现了实时的无人车系统。.本项目取得了丰富的理论成果，积累了重要的实验数据，培养了多名硕士和博士研究生。基于项目科研成果，项目组已在知名学术刊物和会议上发表论文16篇，申请发明专利6项。