Automatic extraction of road curbs from uneven, unorganized, noisy and massive 3D point clouds is a challenging task. Existing methods often project 3D point clouds onto 2D planes to extract curbs. However, the projection causes loss of 3D information which degrades the performance of the detection. This paper presents a robust, accurate and efficient method to extract road curbs from 3D mobile LiDAR point clouds. Our method consists of two steps: 1) extracting the candidate points of curbs based on the proposed novel energy function and 2) refining the candidate points using the proposed least cost path model. We evaluated our method on a large-scale of residential area (16.7GB, 300 million points) and an urban area (1.07GB, 20 million points) mobile LiDAR point clouds. Results indicate that the proposed method is superior to the state-of-the-art methods in terms of robustness, accuracy and efficiency. The proposed curb extraction method achieved a completeness of 78.62% and a correctness of 83.29%. These experiments demonstrate that the proposed method is a promising solution to extract road curbs from mobile LiDAR point clouds.