无人驾驶车辆解耦混合运动规划方法研究

高速公路无人驾驶车辆运动规划要充分考虑车辆的动力学特性和道路环境构成的复杂约束，优先确保车辆的无碰撞行驶轨迹，同时算法实时性比常速情况下要求更高，要充分考虑计算的复杂性。提出一种可生成横纵向最优解逼近群的解耦增强混合运动规划方法，引入Frent坐标系将运动规划问题进行横纵向解耦，在横向偏移规划中融合数值优化和多项式规划算法，在纵向速度规划中融合围绕速度和围绕避障规划算法，解耦规划完成后使用代价函数和碰撞检测方法对生成的轨迹群进行评估和选择。搭建Apollo-LGsvl无人驾驶联合仿真测试平台，对所提出的运动规划方法进行仿真验证。结果表明，提出的横纵向解耦增强混合运动规划算法在高速公路各工况中所生成的运动轨迹能够有效避免车辆发生碰撞，其横向偏移和纵向速度均能满足二阶平滑，最大横向加速度、最大纵向加速度、最大纵向jerk等满足动力学约束。该方法有效减少了最优解逼近群内的轨迹数量，使轨迹评估阶段计算复杂度降低了46%，证明提出的方法具备一定的应用价值。

Research on Decoupling Hybrid Motion Planning Method for Driverless Vehicles

The motion planning of unmanned vehicles on expressway should fully consider the vehicle dynamic characteristics and complex constraints due to traffic flow based on collision-free trajectories. The higher vehicle speed requires better real-time performance of the algorithm and involves more complex calculations. A hybrid motion planning method with decoupled transverse and longitudinal enhancement is proposed to generate the optimal approximation group. The Frent frame is introduced to decouple the motion planning problem laterally and longitudinally. The numerical optimization and the polynomial planning algorithm are fused in lateral motion planning. The planning considering velocity and the planning considering obstacle avoidance are fused in longitudinal planning. Once the decoupled planning is completed,the resulting trajectory groups are evaluated and selected using the cost functions and the collision detection method. An Apollo-LGsvl driverless co-simulation testbed is built to simulate and validate the proposed motion planning method. The experimental results show that the motion trajectories generated by the proposed hybrid motion planning algorithm can effectively avoid vehicle collisions under every set of experimental conditions on the highway. The transverse offset and longitudinal velocity satisfy the second order smoothing condition. The maximum transverse acceleration, the maximum longitudinal acceleration and the maximum longitudinal jerk meet the dynamics constraints. Meanwhile the number of trajectories in the optimal approximation group is decreased effectively, and the time complexity of the trajectory evaluation algorithm is reduced by 46%.