煤矿救援机器人在非结构化井下巷道环境中执行搜救任务，传统的路径规划算法用于搜索空间较大或者复杂度较高的非结构化环境时，可能会面临效率低、非最优路径和平滑度较差等问题；此外，巷道存在复杂的交叉口等环境特征，机器人在该区域容易偏离预设路线或与路口壁面发生剐蹭。为了克服这些问题并提高机器人的导航精度，提出了针对轮式煤矿救援机器人的路径规划算法改进方案：①改进启发式全局路径规划A*算法，通过分层邻域搜索与剪枝的方法改进邻域搜索方式，优化代价函数以更好地平衡实际代价和启发式代价的影响，更准确地评估每个节点的代价，使其适应实际情况且缩减了计算量，并利用B样条方法对路径进行平滑处理；②利用随机样本一致算法（RANSAC）构建煤矿巷道壁面的几何模型，以便提取交叉路口特征点坐标纳入规划系统，利用B样条曲线基函数的局部支撑性优化路径，当后续添加路径优化点时，仅影响相应区间的曲线形状，对路径的其余部分不产生影响；③根据构建的环境几何模型和提取的特征点建立综合局部势场，引入调整系数优化势场分布，并使用粒子群算法（PSO）优化的PID算法实现运动控制，增强机器人针对巷道交叉口等复杂环境通行的适应能力。通过MATLAB和机器人操作系统（ROS）仿真验证了算法原理与应用的可行性，结果表明该方法可以实现在复杂交叉环境中路径规划、自主行驶等功能。

Coal mine rescue robots perform search and rescue tasks in unstructured underground tunnel environments. Traditional path planning algorithms may encounter issues such as low efficiency, non-optimal paths, and poor smoothness when applied to search spaces that are large or complex. Additionally, tunnels feature complex environmental characteristics such as intersections, where robots are prone to deviating from preset routes or scraping against tunnel walls. To address these challenges and enhance the navigation accuracy of robots, improvements to the path planning algorithm for wheeled coal mine rescue robots are proposed: ① The heuristic global path planning A* algorithm is enhanced by employing layered neighborhood search and pruning techniques to optimize the search process. The cost function is refined to better balance the influence of actual cost and heuristic cost, thus more accurately assessing the cost of each node, adapting to real situations, reducing computational complexity, and smoothing the path using B-spline methods. ② The Random Sample Consensus (RANSAC) fitting algorithm is utilized to construct a geometric model of coal mine tunnel walls, facilitating the extraction of feature point coordinates of intersections for inclusion in the planning system. The path is optimized using the local support property of B-spline basis functions. When additional path optimization points are added subsequently, only the shape of the curve in the corresponding interval is affected, leaving the rest of the path unaffected. ③ A comprehensive local force field is established based on the constructed environmental geometric model and extracted feature points. Adjustment coefficients are introduced to optimize the distribution of the force field, and motion control is achieved using the Particle Swarm Optimization (PSO) optimized PID (Proportion Integral Differential) algorithm, enhancing the robot's adaptability to complex environments such as tunnel intersections. The feasibility of the algorithm principles and applications is validated through MATLAB and ROS (Robot Operating System) simulations. Experimental results demonstrate that the proposed method can realize functions such as path planning and autonomous driving in complex intersection environments.