Abstract: To address the low path planning efficiency of mining vehicles in narrow, winding underground tunnels with unknown obstacles, a global path planning algorithm, DVGA^*, was proposed, integrating simplified visibility graphs (SVG) and the A^* algorithm. Based on the construction of a point cloud map of the real environment, the algorithm connected the vehicle's visual tangent points from different viewpoints to dynamically generate the SVG. The visual tangent points were sequentially stored in the OPEN list as nodes, and nodes were selected for the CLOSED list based on the A^* algorithm's evaluation function to ensure the shortest path. This process continued until the endpoint appeared in the OPEN list, resulting in the optimal path points being stored while the remaining nodes in the OPEN list were deleted. Finally, a path smoothing algorithm was utilized to further reduce the number of path nodes, thereby enhancing path planning efficiency. Experimental results indicated that compared to the Complete Visibility Graph + A^* algorithm, SVG + A^* algorithm, and SVGCA^* algorithm, the DVGA^* algorithm had the shortest planning time for complex long-distance paths, with average path lengths reduced by 10.79%, 6.26%, and 2.86%, respectively, demonstrating stronger adaptability and higher planning success rates. Results from underground tests showed that in areas with variable tunnel widths and while avoiding static obstacles, the path planned by DVGA^* was smoother compared to that of the SVGCA^* algorithm. When avoiding dynamic obstacles, DVGA^* was able to promptly correct the path, ensuring timely and stable path planning. In complex and variable tunnel environments, the planning time and path length of DVGA^* were reduced by 11.51% and 1.54%, respectively, compared to SVGCA^*, indicating higher environmental adaptability and stability.