复杂避障约束下自主驾驶轨迹优化

发布时间：2018-06-16 10:19

本文选题：自主泊车 + 时空分割　；参考：《浙江大学》2016年硕士论文

【摘要】：和人类驾驶员相比,无人车能够更加全面的掌握即时路况并及时对汽车巡航状态进行调整,从而改善交通拥堵状况、避免了人为失误造成的交通事故和伤亡,因此近些年来无人驾驶技术发展迅速。许多半自动驾驶技术在汽车上已经得到了大规模的普及,比如紧急制动,定速巡航和车道保持等。然而,在行车环境建模、避障轨迹优化等方面还需要深入的研究。如车载传感器精度有限的情况下如何对障碍环境建模,如何处理环境中意外出现的动态障碍物,如何针对不同的泊车位实现标准化的轨迹优化算法设计,如何优化智能无信号灯路口下的多车避障轨迹等都是值得研究的问题。本文用动态优化全联立算法对上述问题做了一些研究。主要内容和成果如下：1.对城市环境下的自主泊车问题,采用MPCC和R函数方法对车位环境建模,与车辆运动学模型、物理约束共同构成了行车系统模型,构造了联立框架下的自主泊车动态优化命题。采用有限元正交配置法将原命题离散化为非线性数学规划问题,由非线性求解器高效求解得到具有时间信息的可直接用于指导车辆跟踪的泊车轨迹。2.针对自主泊车轨迹动态优化命题含有较多复杂约束可能引起的求解困难,提出了时空分割策略来增强优化算法的收敛性。通过在轨迹优化命题中引入吸引区、塌缩区来分割泊车空间,将非线性的复杂环境约束在割裂空间下进行简化,重构泊车轨迹优化命题。仿真实验证明了时空分割策略的有效性。3.在城市环境下基于信息完整假设进行多车轨迹优化的全局规划算法研究。在多车模型、环境模型下融合了车-车、车与动态可预测障碍物的复杂避障约束,构造多车协作避让轨迹优化命题。数值实验表明了基于全联立的全局规划算法的有效性。4.对于环境感知不完整的车辆轨迹规划问题,基于障碍环境预测模型进行局部滚动优化。运用假设静态法、速度切线预测法、完整预测法对障碍环境建模,根据障碍车辆进出我车的冲突检测域来切换重构行车系统轨迹优化命题,并比较了预测模型对车辆避障性能的影响。
[Abstract]:Compared with human drivers, the UAV can master the real-time traffic conditions more comprehensively and adjust the vehicle cruising state in time, thus improving the traffic congestion and avoiding the traffic accidents and casualties caused by human error. As a result, driverless technology has developed rapidly in recent years. Many semi-autonomous driving techniques have been widely used in automobiles, such as emergency braking, constant speed cruising and lane maintenance. However, further research is needed in traffic environment modeling and obstacle avoidance trajectory optimization. For example, how to model the obstacle environment, how to deal with the unexpected dynamic obstacles, how to design the standardized trajectory optimization algorithm for different parking spaces, how to model the obstacle environment under the condition of limited precision of the vehicle sensor, how to deal with the unexpected dynamic obstacles in the environment, It is worth studying how to optimize the trajectory of multi-vehicle obstacle avoidance at the intersection of intelligent signal-free. In this paper, the dynamic optimization algorithm is used to study the above problems. The main contents and results are as follows: 1. For the problem of autonomous parking in urban environment, the vehicle parking environment is modeled by MPCC and R function method, and the vehicle kinematics model and physical constraints are combined to form the vehicle system model, and the dynamic optimization proposition of autonomous parking under the simultaneous frame is constructed. The finite element orthogonal collocation method is used to discretize the original proposition into a nonlinear mathematical programming problem. The nonlinear solver is used to efficiently solve the parking trajectory with time information which can be directly used to guide the vehicle tracking. In view of the difficulty of solving the dynamic optimization proposition of autonomous parking trajectory with more complex constraints, a spatio-temporal segmentation strategy is proposed to enhance the convergence of the optimization algorithm. By introducing attraction region and collapsing area into the trajectory optimization proposition, the parking space is separated, and the nonlinear complex environment constraint is simplified in the split space, and the parking trajectory optimization proposition is reconstructed. The simulation results show that the spatio-temporal segmentation strategy is effective. The global planning algorithm for multi-vehicle trajectory optimization based on the assumption of information integrity in urban environment is studied. Under the multi-vehicle model and environment model, the complex obstacle avoidance constraints of vehicle-vehicle, vehicle-vehicle and dynamic predictable obstacles are combined, and the proposition of multi-vehicle cooperative avoidance trajectory optimization is constructed. Numerical experiments show the effectiveness of the global programming algorithm based on full synchronization. 4. 4. For the vehicle trajectory planning problem with incomplete environmental perception, the local rolling optimization based on the obstacle environment prediction model is carried out. Using the hypothesis static method, the velocity tangent prediction method, the complete forecast method to model the obstacle environment, according to the obstacle vehicle entering and leaving our vehicle conflict detection domain to switch the reconstruction train system trajectory optimization proposition, The effect of prediction model on vehicle obstacle avoidance performance is compared.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U463.6

【参考文献】