Concepts for autonomous navigation of underground mine face haulage equipment using depth cameras

Abstract

This paper describes the development of a software pipeline for autonomous navigation of underground mine haulage equipment. The approach uses depth-sensing cameras mounted on the haulage vehicle to capture visual and depth information. Semantic segmentation is used to identify various objects in the scene, e.g., roof, ribs, floor, people, and mining equipment, and the depth data is used to develop a segmented point cloud. This information is combined to develop a two-dimensional occupancy grid. The occupancy grid and haulage vehicle position within that map are used by a path planner module to plan a path for the haulage vehicle to the target. The direction and distance information are used by the controller to generate steering and distance signals to execute the path. The methodology is applied to coal mine room-and-pillar face haulage because of the unique challenges this application presents, namely a large haulage vehicle in confined spaces and a constantly changing target location. Development and testing are conducted in a 1/6th -scale mock mine with an autonomously navigated shuttle car and in a simulated mine environment using full-scale mining equipment with autonomous path planning and trajectory signals given to a shuttle car operator.