Abstract: Aiming at the problems of unsmooth, unsafe, and slow generation of the generated path in the traditional path planning algorithm, Bi-RRT was used to study the path planning algorithm in unmanned driving. Firstly, the vehicle’s own steering constraints was used to construct a node expansion space that met the vehicle kinematics requirements. Then, the target bias sampling strategy and two-way search were used to greatly improve the planning speed of the RRT algorithm. At the same time, the KD tree was used to define the nearest neighbor point in combination with the maximum curvature constraint. Finally, the generated path was pruned and subjected to safety collision detection and Bezier curve smoothing to make it more in line with the dynamic characteristics of the vehicle. The research results show that the improved Bi-RRT algorithm has a significant improvement in search speed and algorithm overhead, and the generated path is also smoother, which proves the effectiveness, practicability and adaptability of the proposed algorithm.

Key words: intelligent vehicle; route planning; steering constraints; nearest nodes

摘要： 针对传统路径规划算法中所生成路径不平滑、不安全、生成速度慢等问题，利用双向快速搜索随机树（Bi-RRT）对无人驾驶中的路径规划算法进行了研究。首先利用车辆自身的转向约束构建满足车辆运动学要求的节点扩展空间，随后采用目标偏置采样策略以及双向搜索大大提高了RRT算法的规划速度，同时使用K-D树结合最大曲率约束定义最邻近点，最后对生成的路径剪枝并进行安全碰撞检测和贝塞尔曲线平滑处理，使之更符合车辆的动力学特性。研究结果表明：改进后的Bi-RRT算法的在搜索速度和算法开销上都有显著的提升，生成的路径也更加平缓，证明了该算法的有效性、实用性和适应性。

关键词: 智能汽车；路径规划；转向约束；最邻近点