【作者】 辛煜；

【导师】 梁华为；

【作者基本信息】 中国科学技术大学 ， 检测技术与自动化装置， 2014， 博士

【摘要】 无人驾驶车辆是一种轮式移动机器人,其技术涉及认知科学、人工智能、机器人技术与车辆工程等交叉学科,是验证各种新兴技术的最佳实验平台,也是当今前沿科技的重要发展方向。它既包括理论方法与关键技术的突破,也涉及到大量的工程与试验问题。在无人驾驶车辆的单元技术中,室外复杂环境下的运动障碍物检测、预测和避撞一直是研究的重点和难点。针对无人驾驶车辆在室外复杂环境中进行运动障碍物检测、预测和避撞所遇到的问题以及系统的总体设计要求,本文设计了一种无人驾驶车辆运动障碍物避撞系统,对其中检测、预测和避撞方法进行了深入研究,介绍了在工程应用和实验测试中具体的一些技术手段。具体的研究内容包括以下几个方面：1)对无人驾驶车辆运动障碍物避撞系统的相关概念和国内外研究成果进行了分析,了解无人驾驶车辆运动障碍物检测、预测和避撞的基本流程和基本方法。2)介绍了“智能先锋”号无人驾驶车辆的系统架构和基本方法,分析了室外复杂环境下无人驾驶车辆运动障碍物检测、预测和避撞过程中的关键性问题,提出了系统的总体设计要求,并设计了“智能先锋”号无人驾驶车辆运动障碍物避撞系统。3)研究了无人驾驶车辆运动障碍物检测方法。采用四线激光雷达Ibeo和三维激光雷达Velodyne作为运动障碍物的检测传感器,分别对其数据进行一系列处理,最后融合两者数据得到障碍物运动信息,该方法既能满足无人驾驶车辆对运动障碍物检测准确性的要求,又能满足实时性的要求。4)研究了无人驾驶车辆运动障碍物碰撞预测及表示方法。提出一种新的障碍物占用栅格图——时空障碍物栅格图,它不仅能表示静态障碍物的占用信息,还能将无人驾驶车辆在未来一段时间内与运动障碍物之间的预测碰撞信息表示出来。采用不同的模型来预测不同场景下无人驾驶车辆与运动障碍物之间的碰撞关系,并介绍了时空障碍物栅格图的直接生成方法和优化生成方法。它能够将运动障碍物的避撞与静态障碍物避撞统一起来,简化避撞过程的算法复杂性。5)研究了无人驾驶车辆运动障碍物避撞方法。提出一种可搜索连续邻域的改进A*算法,对栅格进行线性插值,将传统A*算法的可搜索邻域从离散的8个增加为无限个,搜索方向也变为连续的任意方向,能够解决传统A*算法在栅格图中求解出的最短路径并非最优的问题。然后介绍了一种基于这种改进A*算法的无人驾驶车辆动静态障碍物避撞方法,实现无人驾驶车辆在动态环境中安全、智能的自主驾驶。“智能先锋”号无人驾驶车辆平台应用了以上研究内容,对其进行了实车测试并参加了由国家自然科学基金委举办的“智能车未来挑战赛”,实验结果和优异的比赛成绩验证了系统及其具体方法的有效性和可行性。本文最后总结了该系统中存在的优势和不足,也对进一步的研究提出了展望。更多还原

【Abstract】 The autonomous vehicle is one kind of wheeled mobile robots, whose technologies are related to cognitive science, artificial intelligence, robot technology, automotive engineering, and many other interdisciplines. It is the best experimental platform to exam the newly developing technology and an important development direction of today’s advanced science and technology. The autonomous vehicle not only includes the breakthrough of theoretical approaches and key technology, but also, it is related to the issue of engineering and experimentations.Of the unit technology of autonomous vehicles, dynamic obstacles detection, prediction and avoidance in complex outdoor environments is always the emphasis and difficulty of the research. On account of the problem autonomous vehicles encountered in research on dynamic obstacles detection, prediction and collision avoidance in complex outdoor environments, and the general design requirement of the system, a dynamic obstacles avoidance system of autonomous vehicles is presented. In this paper, the methods in detection, prediction and avoidance are researched indepth and some specific techniques of engineering application and experimental testing are introduced. The specific research contents include some aspects as follows.1) The related notion and research results abroad and in China of dynamic obstacles avoidance system of autonomous vehicles are analysed. The basic flow and basic approach of dynamic obstacles detection, prediction and avoidance of autonomous vehicles are understood.2) The system architecture and basic approach of "Intelligent Pioneer" autonomous vehicles are presented. The critical problem of dynamic obstacles detection, prediction and avoidance of the autonomous vehicle in complex outdoor environments is analysed and the general design requirement of the system is proposed. In addition, the "Intelligent Pioneer" dynamic obstacles avoidance system of autonomous vehicles is designed.3) Dynamic obstacle detection approach for the autonomous vehicle is researched. The four layers laser radar "Ibeo" and the three-dimensional laser radar "Velodyne" are used as detection sensors for dynamic obstacles. Their data are respectively processed through a series of processes. Finally, motion information of dynamic obstacles processed from these two types of data is fused. This approach not only meets the demand of accuracy, but also, it meets the demand of real-time of dynamic obstacle detection of autonomous vehicles.4) Dynamic obstacle collision prediction and representation approach for the autonomous vehicle is researched. In this paper, a new occupancy grid named space-time occupancy grid map is presented. It can not only represent the occupancy information of static obstacles, but also, it can represent the predicted collision information of the autonomous vehicle and dynamic obstacles in a period of future time. Different models are used to predict the collision between the autonomous vehicle and dynamic obstacles when the autonomous vehicle travels in different scenarios. The directly generating method and the optimization generating method of space-time occupancy grid map are introduced. This grid map unites the collision avoidance of dynamic obstacles and static obstacles, and simplifies the algorithm complexity of the collision avoidance process.5) Dynamic obstacle avoidance approach for the autonomous vehicle is researched. In this paper, an improved A*algorithm for searching consecutive neighbourhoods is presented. Linear interpolation of the grid is executed. As a result, the searchable neighbourhood number of the conventional A*algorithm is changed from8to infinite and the searchable direction is with an arbitrary and continuous degree. It can solve the problem that the optimal path calculated by the conventional A*algorithm on grid map is not the best one. Then a dynamic and static obstacles collision avoidance approach for the autonomous vehicle based on this improved A*algorithm and the space-time occupancy grid map is introduced. This approach can realize secure and intelligent autonomous driving of the autonomous vehicle in dynamic environments.The above research contents are applied on the "Intelligent Pioneer" autonomous vehicle platforms. The "Intelligent Pioneers" have tested on the actual road and participated in the "Intelligent Vehicle Future Challenge of China" Competition funded by Natural Science Foundation of China. The experimental result and excellent performance in the competition validate the effectiveness and feasibility of this system and the specific methods. At the end of this paper, the advantages and disadvantages of this system are summarized, and an expectation of further research is brought out.更多还原