Abstract:Purpose: This paper aims to overcome the edge-hugging, corner-cutting, and poor smoothness that the classical A* algorithm exhibits in underground roadway navigation by proposing an enhanced variant tailored to narrow and winding underground roadways, thereby improving mobile-robot safety and path quality. Methods: This paper augments the conventional A* framework with a diagonal collision-avoidance check to eliminate corner cutting. An underground-roadway-width factor is embedded in the heuristic to guide the search toward the centerline, enhancing path stability and feasibility. Furthermore, path length, mean curvature, fluctuation degree, and heading-change rate jointly determine the order and smoothing factor of a B-spline curve for post-processing. Performance before and after the improvement is assessed through comprehensive simulations and field tests. Results: The improved algorithm automatically keeps the path centered and produces smooth turns. Compared with the classical A*, node expansions are reduced by 15.58%, computation time is shortened by 34.37%, and the number of way-point corners decreases by 40%, effectively mitigating edge-hugging, sharp turns, and unsmooth path. Significance: This paper demonstrates that the proposed algorithm markedly improves path smoothness, central alignment, and computational efficiency, enabling safer and more reliable robot traversal in complex underground roadways. The approach offers substantial benefits for underground transport, inspection, and rescue tasks and provides a solid algorithmic foundation for future intelligent-system deployments.