Abstract: At present, the positioning technology for mobile robots in coal mine underground environments still cannot meet the requirements for large-scale high-precision positioning, and it fails to achieve automatic alignment between positioning results and the absolute coordinate system of the underground coal mine. The key bottleneck is the lack of an effective regional absolute positioning method to intermittently correct the cumulative errors from long-term robot operation. In light of this, to achieve local high-precision positioning of coal mine mobile robots within the absolute coordinate system, an absolute positioning method is proposed based on roadway-side LiDAR 3D object detection. By deploying fixed LiDAR sensing nodes at key locations, it provides critical technical support for constructing a vehicle-road cooperative positioning system framework. First, based on the roadway environment and LiDAR’s operational characteristics, the selection and installation design of LiDAR are conducted, and a method for calibrating the extrinsic parameters between the LiDAR coordinate system and the absolute roadway coordinate system is proposed. Second, to address degraded positioning accuracy due to sparse point clouds of distant robots, an attention mechanism is introduced into the PointPillars-based 3D object detection network. Through feature enhancement in both channel and spatial dimensions, the model's ability to identify and locate sparse, weak-feature targets is effectively improved. Then, comparative experiments on the public KITTI dataset demonstrate that the improved strategy significantly outperforms the original baseline, validating the superiority of the proposed attention mechanism in enhancing detection accuracy. Finally, field experiments in a simulated roadway environment verify the stability, robustness, and real-time performance of the absolute positioning method based on roadway-side LiDAR 3D object detection. The ground truth of robot positioning measured by a total station is compared with the algorithm estimates. The results show that the absolute positioning algorithm achieves an average error of 0.167 m and an average runtime of 0.062 s while meeting the requirements of robustness and real-time performance. The proposed method satisfies the need for absolute positioning of underground robots and can provide intermittent absolute positioning constraints for odometry-based positioning of coal mine mobile robots over long distances and large scenes.