To enhance terrain adaptability and storage efficiency in mobile robot path planning on non-flat terrains, an improved A* path planning algorithm based on the dual-resolution hierarchical grid map is proposed. This map includes a high-resolution obstacle layer using binary representation for rigid obstacles and a low-resolution elevation layer quantifying terrain undulations via a digital elevation model. On this basis, the A* algorithm is improved by reconstructing its dynamic weighted composite cost function. The improved algorithm introduces three optimization dimensions into the mobility cost function, including slope constraints, energy consumption weight, and safety factor. The heuristic function is extended to a multimodal evaluation metric that integrates spatial distance, root mean square slope, and terrain risk values. A distance-sensitive dynamic weight adjustment strategy is designed, and the Sigmoid function is utilized to achieve a smooth transition between global heuristic search and local path optimization. Experiments show that within a rectangular mapping range of 700 m×700 m, the dual-resolution hierarchical grid map structure reduces storage load by 61.7% compared with a three-dimensional grid map. Compared with traditional A* algorithms, this method reduces the standard deviation of elevation fluctuations in planned paths by 38.9%. Real robot experiments demonstrate that this method effectively avoids steep slopes and obstacles. Engineering application experiments indicate that this method reduces memory usage by 62% in large-scale unstructured scenarios such as oilfield inspections, with path planning response times under 6.9 s and the planned paths exhibiting gentle low undulation characteristics.