基于采样优化与双向梯度的两阶段轨迹规划方法

Two-Stage Trajectory Planning with Sampling Optimization and Bidirectional Gradient

摘要: 轨迹规划是无人驾驶车辆安全导航的关键环节，其目的是在动态变化的环境中实时生成无碰撞、高效的轨迹。为提高无人驾驶车辆轨迹规划的计算效率和生成轨迹的质量，本文提出了一种基于采样空间优化粗搜索和双向梯度细优化的粗到细两阶段轨迹规划方法。该方法首先根据不同行驶工况生成对应的采样空间，进而在生成的离散化粗全局空间中进行高效搜索，以快速探索可能的轨迹；进一步在连续的细化局部空间中进行双向梯度下降优化，以精细化调整轨迹，降低轨迹的成本。通过这种两阶段的方法，不仅能够加快规划速度，还能生成更优的轨迹，有效提升了轨迹规划的性能。实验表明，该方法对比基线运行时间减少了16.1%、轨迹总代价降低了12.6%。

Abstract: Trajectory planning is a critical component for the safe navigation of autonomous vehicles, aiming to generate collision-free and efficient trajectories in real-time within dynamically changing environments. To improve the computational efficiency of trajectory planning and the quality of generated trajectories, this paper proposes a coarse-to-fine two-stage planning method based on sampling-space-optimized coarse search and bidirectional gradient fine optimization. The method first generates adaptive sampling spaces according to different driving scenarios, enabling efficient exploration of possible trajectories through discretized coarse global search. Then, it performs bidirectional gradient descent optimization in a continuous refined local space to fine-tune the trajectory and minimize its cost. By leveraging this two-stage approach, the method not only accelerates planning speed but also produces higher-quality trajectories, significantly enhancing overall planning performance. Experiments demonstrate that compared to baseline methods, this approach reduces runtime by 16.1% and decreases total trajectory cost by 12.6%.