针对狭窄空间中路径规划效率低、成功率低及碰撞检测耗时占比高的问题，我们提出一种基于高维构型空间快速碰撞检测的路径规划算法。该算法以RRT-Connect为基线搜索框架，并行运行快速碰撞检测模块：首先通过启发式策略均衡采集狭窄通道内的自由与碰撞构型，构建高质量数据集，克服均匀采样获取的数据集中构型数量不均衡的问题；随后对两类构型在线聚类，以簇的形式表征构型空间分布，从而将传统基于包围盒的碰撞检测转化为采样构型与聚类簇间的距离计算，大幅降低单次检测的耗时。我们在简单、开放、封闭三类狭窄环境下的仿真与实验结果表明，该算法在路径搜索效率和成功率上均表现优异，为机器人在复杂受限场景中的规划提供了高效可靠的解决方案。
赖希睿, 王润花*, 王耀南, 张雪波, 杨磊, 李帅. 狭窄空间下基于高维构型空间快速碰撞检测的机械臂在线路径规划. 自动化学报（已录用）
Abstract
Aiming at the problems of low efficiency and success rate, high percentage of time-consuming collision detection in robotic arm path planning within narrow spaces, this paper proposes an online path planning algorithm for narrow spaces utilizing online fast collision detection in high-dimensional configuration space. The algorithm takes RRT-Connect as the baseline path search framework, and runs the fast collision detection module based on high-dimensional configurations online clustering in parallel. The module includes balanced sampling of the high-dimensional configuration dataset and online training of the fast collision detection model. Specifically, in the phase of dataset online construction, a heuristic strategy is introduced to fully explore the free configurations in narrow space, overcoming the problem that uniform sampling leads to an uneven number of collision and free configurations in the dataset, and to provide effective data support for the subsequent model training; After the dataset construction, the distribution of the two types of configurations in the high-dimensional space is characterized in the form of clusters by online clustering of collision and free configurations; Using the trained cluster model, the collision detection based on bounding box in the baseline algorithm can be transformed into the calculation of the distance between the sampled configuration and the cluster, which greatly reduces the time consuming for a single collision detection and effectively improves the efficiency of proposed algorithm. Through simulation tests and experimental verification in simple, opened and closed three types of narrow environment, the results show that the proposed algorithm has significant advantages in terms of path searching efficiency and success rate.
