利用激光点云的城市无人驾驶路径规划算法

为了解决随机采样算法受感知环境不确定性影响下的弱鲁棒性以及弱可靠性问题，采用一种基于激光空间势场的渐优随机采样算法框架来设计符合无人驾驶需要的规划算法。针对感知环境的不确定性，首先基于势场原理与激光障碍物点云构建一个融入了斥力场的规划空间，解决激光障碍物提取中的过分割等问题。其次，利用规划空间来处理随机采样算法中的采样策略、最优母节点选取策略、修剪策略以及最终路径选择策略。再次，在算法中加入了Anytime策略来提高优化解的利用率，使得算法的计算效率满足无人驾驶实时性的要求。同时，为了保证无人驾驶中规划路径的鲁棒性与可靠性，创建了一个综合5重因素的代价函数来选择最优路径，并根据不同的无人驾驶场景来调整相对应的参数；最后在城市测试道路上进行了实地测试。结果表明：设计的算法框架能够适应最高时速40 km·h^-1的城区驾驶环境，并能完成跟驰、换道、融入以及静动态障碍物的避障决策。在与SST算法的对比试验中，所提出的算法在各个试验中的轨迹、方向盘转角以及速度的平滑性都优于SST算法，其轨迹与障碍物的距离也优于SST算法。

Path Planning of Urban Autonomous Driving Using Laser Point Cloud Data

To solve the problems of weak robustness and low reliability of random sampling algorithm under the influence of perceived environmental uncertainty, a gradual optimal sampling algorithm framework based on laser obstacle potential space field was adopted in designing the planning algorithm for autonomous driving requirements. First, a planning space that integrates the repulsive field based on the potential field and laser point cloud was constructed to solve the over-segmentation problem in laser obstacle extraction. Second, the planning space was used to process the sampling, optimal parent node selection, pruning, and final trajectory selection strategies in the random sampling algorithm. Third, the “anytime” strategy was added to the algorithm to improve the utilization of the optimized solution, and the algorithm's calculation process met the requirements of the real-time autonomous driving system. Fourth, to improve the robustness and reliability of autonomous driving, a cost function that combines the effects of the above planning space and five factors for selecting the optimal path was created, which was based on different autonomous driving scenes to adjust the corresponding parameters. Fifth, a real autonomous driving car was used to conduct field tests on an unmanned city road, including keeping and changing lanes, overtaking, and static obstacle avoidance. The test results show that the algorithm framework designed in this study can adapt to the urban environment driving conditions with a top speed of 40 km·h^-1, and can complete the car-following, lane changing, integration, intersection driving, and obstacle avoidance decisions of static and dynamic obstacles. In comparison with the SST algorithm, the smoothness of the trajectory, steering wheel angle, and speed of the algorithm in all the experiments are superior to those of the SST algorithm. Additionally, the distance between the trajectory and obstacle is also better than that of the SST algorithm.