面向开放式无人驾驶车群的正则性自协机理研究

There objectively exist some defects on the complex relationship of high-dimensional data, low coupling on interconnection, real-time and stability that can’t be guaranteed in the open self-driving vehicle group, which cann’t effectively support the information fidelity and the implementation faithfulness within the self-driving group, that is the regularity problem, and it has become an obstacle to large-scale applications and industrial development. The current researches mainly focused on the independent decision-making of single agents and self-association of closed self-driving vehicle group. It lacks theoretical principle and methods of self-association of self-driving vehicle group driving behavior in an open road network environment. This proposal breaks through the current research idea in the small-scaled closed-loop system, and considering the open road network environment. We propose a theoretical system on "data analysis→behavior evolution→model self-association". The research contents are as following: First, we introduce deep learning to analyze the complex relationship of high-dimensional data, and propose a small data analytic model to rapidly construct self-driving vehicle group in real scene; Second, with the aid of Pareto optimal theory, the selection model of leading node are given; Third, combined with swarm intelligence, the dynamic evolving events and conversion method of self-driving and its complete self-association theory model are proposed; Finally, the optimization mechanisms on self-association mechanism are given. After solved the above scientific problems, the theoretical system and methods which can support intelligent movement of open self-driving vehicle will be established, and it could also explore new theories and methods for the key common technologies in artificial intelligence.

开放式无人驾驶车群客观存在高维数据复杂关系、互连互通耦合度低、实时性和稳定性无法保证等缺陷，难以有效支撑车群内部实现信息保真和执行忠实，即正则性问题，成为阻碍无人驾驶大规模应用和产业发展的瓶颈。现有研究主要集中于单智能体独立决策以及封闭式无人驾驶车群自协，缺乏开放式路网环境下无人驾驶车群运动行为自协理论和方法。本课题突破现有小尺度闭环系统的研究思路，从开放式路网环境出发，提出“数据解析→行为演化→模型自协”的理论体系。具体包括：引入深度学习解析高维数据复杂关系，提出实际场景下快速构建无人驾驶车群的小数据解析模型；借助帕累托最优理论，给出引领节点选择模型；结合群体智能思想，提出无人驾驶车群动态演化事件和转换方法及其完备性自协理论模型；给出自协机理优化机制。解决上述科学问题，建立一套支持开放式无人驾驶车群运动行为智能化的理论体系和方法，也为人工智能基础理论和关键共性技术体系探索新的理论和方法。

本课题突破现有小尺度闭环系统的研究思路，从开放式路网环境出发，提出“数据解析→车群形成→动态演化→模型自协”的理论体系。主要成果如下：.（1）利用深度学习对数据进行解析，分析无人驾驶车辆之间以及与外部干扰之间的关联关系，建立小数据解析模型，预测车辆之间的网络性能影响因素。.（2）给出了无人驾驶车辆节点的角色分类，引领节点选择，以及车辆连通性，耦合性以及实时性等相关性质；.（3）给出了开放式路网环境下，考虑各种影响因素干扰情况下的无人驾驶车群模型的构建，车群规模以及无人驾驶车群模型相关性质的证明；.（4）分析了有人驾驶、路边障碍物、红绿灯和行人等干扰对无人驾驶车辆之间通信的影响，探讨了城市场景中无人驾驶车群动态演化原因，给出加入、离开、合并、分裂和消亡五类动态演化事件及其处理方法，探寻动态演化事件与车群状态之间的关联关系，进而挖掘出车群动态演化规律，即车群演变性。.（5）给出了无人驾驶车群自协准则：间隔准则，聚合准则，匹配准则。.（6）综合考虑无人驾驶车群的连通性、耦合性、实时性、演变性和自适应性等性质，给出了无人驾驶车群的自协模型，模型的求解，以及自协模型相关性质的证明。.（7）围绕“风险评估效用→车群形成→动态演化→安全自协模型”，给出了安全的无人驾驶车群自协模型。.（8）围绕“侧链共识→无人驾驶车群形成→车群动态演化→车群协同模型”，给出了高速公路场景下无人驾驶车群协同模型。.取得了上述成果，建立一套可支持开放式无人驾驶车群运动行为智能协同的理论体系和方法，为未来无人驾驶智能化发展奠定理论和应用基础。