Research On Behavioral Decision Making And Motion Planning Methods Of Autonomous Vehicle Based On Human Driving Behavior

In recent years, with the sustained growth of the global traffic accident rate and car ownership, traffic safety and congestion have become increasingly serious. And then, the mission of building Intelligent Transportation System is more urgent. Autonomous vehicle, as one of the key points to organize the intelligent transportation system, has become the focus of world attention.Autonomous vehicle belongs to one kind of outdoor mobile robot, also known as robotic vehicle. The autonomous vehicle is a synthetic intelligent system consisted of environmental perception, behavioral decision making, motion planning and autonomous control. And it comes down to machinery, control, sensor technology, signal processing, pattern recognition, artificial intelligence, computer technology and other multi-disciplinary knowledge. Moreover, it includes the research of scientific theory and method, the breakthrough of key technology, and also a large number solutions of engineering practice problems. Therefore, it owns quite important social value in scientific research and practical application.Behavioral decision making subsystem and motion planning subsystem are two important parts of autonomous vehicle system. Recently, they are research hotspots and also intractable issues in the field of autonomous vehicle. In the comprehensive road environment, due to the complicated and fast-changing driving scenarios, the unpredictable traffic participants, and the public’s improvement of the requirement of traveling comfort, efficiency and safety, the existing behavioral decision making and motion planning method cannot find out a reasonable solution. To this end, this paper developed a kind of driving behavior decision-making model based on decision tree through studying the driving behavior decision-making process of human in comprehensive traffic scene. Then, a kind of drivers’visual behavior-guided RRT motion planning method was proposed by further researching driving vision attention mechanism of human in the process of driving. The specific research content as follows:1) Based on studying the development present situation of autonomous vehicle at home and abroad, this paper deeply comprehended the relevant technologies of the behavioral decision-making and motion planning for autonomous vehicle. Thus, the system architectures and construction schemes for these subsystems were also be analyzed and compared.2) The paper introduced the autonomous vehicle platform named "Intelligent Pioneer II" in details. The function of each module and also the mutual collaborative relationships were analyzed and discoursed. Through the study of behavior decision mechanism in the process of driving and the consideration of traveling comfort, efficiency and safety, this paper proposed the key problems of the system design of behavioral decision-making and motion planning. Thus, this study clearly puts forward system design criteria to achieve the construction of behavioral decision-making subsystem and motion planning subsystem.3) Aiming at the diversity and time-variability of driving scenarios and the showed different behavior features of human drivers in different scenarios, this paper adopt the method of finite state machine to build scene-based behavioral decision-making subsystem of autonomous vehicle. Firstly, in order to realize the reasonable transformation of different driving states, this paper put forward a kind of driving scene transformation model based on finite state machine by studying the change regularity of the drivers’behavior in the different driving scenes. From the standpoints of the practical engineering application, this paper present a method of driving behavior decision-making modeling based on ID3 decision tree by studying the related condition attributes, which have directly influences on the driving behavioral decision-making results. This paper adopt a condition attribute analysis method based on grey relation entropy for the sample data sets of human driving experience, to find out the order of grey entropy relation grade of all the condition attributes and confirm their influences. And then, this method can able to build out the compact driving behavior decision tree model with lower redundancy, thus teasing out the production rules of driving behavior decision-making in line with the thinking process of human drivers for decision-making. By aforementioned methods, for the decision-making subsystem, the poor real-time performance and non-accuracy issues, caused by the insufficiency of knowledge acquisition and representation of the multi-source heterogeneous information, can be solved.4) Research on motion planning method of autonomous vehicle based on drivers’ visual behavior. Firstly, this paper illustrated human visual behavior characteristics in the process of driving and then analyzed the inner link of between the visual behavior and motion planning. Secondly, this paper stated the technical difficulties of motion planning for autonomous vehicle. Then, target on the complex and variability of driving environment in comprehensive driving scenarios, a drivers’ visual behavior-guided RRT motion planning method was proposed by combining the Two-point Visual Model, which was developed by behavioral scientists. According to the decision instruction and the change rule of human visual attention in the driving process, to achieve the extraction of visual attention points by effectively fusing the data of camera and 3D laser radar point cloud. And these identified points can guide the sampling of the random extension node of RRT algorithm. Then on this base, a trajectory optimization method with continuous curvature properties by considering the kinematic and dynamic constraints of vehicle was developed. This method can make the resulting trajectory in accordance with human’s behavioral and thus able to more reasonable adapt to the various dynamic road environment.In the end, the effectiveness and feasibility of the proposed behavioral decision-making and motion planning approaches are verified in the real road environment, which is based on the autonomous vehicle platform "Intelligent Pioneer Ⅱ".