针对煤矿救援机器人采用双向A^*算法存在搜索效率低、路径安全性和平滑性差，及动态窗口法（DWA）融合全局路径规划算法存在实时寻路效率低等问题，提出了一种基于分层平滑优化双向A^*引导DWA（HSTA^*−G−DWA）算法的煤矿救援机器人路径规划方法。首先，将碰撞约束函数的调整机制引入双向A^*算法中，以提高路径规划的安全性。其次，在双向A^*算法的代价函数中增加归正因子函数，防止正反向搜索路径不相交的情况，同时为预估代价函数增加动态加权因子函数以剔除路径搜索过程中无关扩展节点的搜索，从而提升路径搜索效率。然后，利用分层平滑优化策略消除路径中的冗余点和转折角，以减少节点数量和路径长度，并提高路径平滑性。最后，若煤矿救援机器人按照初始全局路径行驶过程中探测到未知障碍物，则利用全局路径引导DWA实现局部动态避障。仿真实验结果表明：① 静态环境下HSTA^*−G−DWA算法路径搜索时间较传统A^*算法和双向A^*算法分别平均减少了81.82%和64.63%，路径的安全性和平滑性更好。② 未知环境下HSTA^*−G−DWA算法可实时避开环境中出现的未知障碍物，路径长度较快速扩展随机树（RRT）算法、改进A^*算法和现有融合算法分别减少了10.34%，14.28%和2.45%，路径搜索时间较现有融合算法平均减少了70.48%。实验室环境下实验结果表明：① 静态环境下，HSTA^*−G−DWA算法路径搜索时间较传统A^*算法平均减少了58.75%，机器人边缘距障碍物的最小距离平均增加了0.71 m。② 未知环境下，相比于传统A^*算法，HSTA^*−G−DWA算法可实时避开环境中出现的未知障碍物且路径的平滑性更好。

This paper proposes a path planning method for coal mine rescue robots based on a Hierarchical Smooth Optimization Bidirectional A^* guided dynamic window approach (HSTA^*-G-DWA) algorithm. The method addresses several limitations of the traditional bidirectional A^* algorithm, including low search efficiency, poor path safety, and inadequate smoothness, as well as low real-time pathfinding efficiency when integrating the DWA with global path planning algorithms. Firstly, an adjustment mechanism of collision constraint function is incorporated into the bidirectional A^* algorithm to improve path safety. Next, a correction factor is incorporated into the cost function of the Bidirectional A^* algorithm to ensure that the forward and backward search paths intersect, preventing them from diverging. Additionally, a dynamic weighting factor is added to the estimated cost function to eliminate irrelevant expanded nodes during pathfinding, thus improving search efficiency. A hierarchical smoothing optimization strategy is employed to remove redundant points and sharp turns, reducing both the number of waypoints and the overall path length, while enhancing smoothness. Finally, if the robot detects unknown obstacles while traveling along the global path, the DWA, guided by the global path, enables dynamic local obstacle avoidance. Simulation results show that: ① In static environments, the path search time using the HSTA^*-G-DWA algorithm is reduced by 81.82% and 64.63% on average compared to the traditional A^* and bidirectional A^* algorithms, respectively, with improved path safety and smoothness. ② In unknown environments, the HSTA^*-G-DWA algorithm can avoid unknown obstacles in real time, reducing the path length by 10.34%, 14.28%, and 2.45% compared to the rapidly-exploring random tree (RRT) algorithm, the improved A^* algorithm, and existing integrated algorithms, respectively. The average path search time is reduced by 70.48% compared to existing integrated algorithms. In laboratory environments, experimental results show: ① In static environments, the HSTA^*-G-DWA algorithm reduces the path search time by 58.75% on average compared to the traditional A^* algorithm, and the minimum distance between the robot's edge and obstacles increases by 0.71 m on average. ② In unknown environments, compared to the traditional A^* algorithm, the HSTA^*-G-DWA algorithm can avoid unknown obstacles in real time, resulting in smoother paths.