Autonomous Docking Using Learning-Based Scene Segmentation in Underground Mine Environments

This paper describes a vision-based autonomous docking solution that moves a coalmine shuttle car to the continuous miner in GPS-denied underground environments. The solution adapts and improves state-of-the-art autonomous docking techniques using a RGBD camera specifically in under-ground mine environments. It includes five processing modules: scene segmentation, segmented point-cloud generation, occupancy grid estimation, path planner, and controller. A two-stage approach is developed to train the scene segmentation network for adapting to the changes from normal environments to dark mines. The resulting network detects both the continuous miner and other objects accurately in mines. Based upon these recognized objects, a path is planned for moving the shuttle car from its initial position to the continuous miner, while avoiding obstacles and other workers. Experiments are conducted using the system in a 1/6th-scale lab environment and data collected in a full-scale realistic mine environment with full-size equipment. The results show the potential of this solution, which can significantly enhance the safety of workers in mining operations.