Autonomous Navigation and Mapping in Underground Mines Using Aerial Robots

In this work we present an integrated approach for autonomous navigation and mapping in underground mines using aerial robots. Underground mines present a set of critical challenges as they are not only GPS-denied by nature, but their environmental circumstances also lead to severe sensor degradation ( due to combinations of darkness, dust, and smoke), localizability problems ( due to textureless surfaces and locally self-similar structure), while also presenting stringent navigation conditions as a result of certain very narrow geometries. To address these issues, the presented robotic systems integrate a multi-modal sensing suite and a set of associated fusion algorithms that enable simultaneous localization and mapping in the GPS-denied and visually-degraded environment of underground mines. More specifically, the fusion of visual and thermal cameras, LiDAR, as well as Inertial Measurement Unit cues is investigated in order to provide informative data in textureless, dark, and obscurants-filled settings. Provided the capacity for reliable pose estimation and mapping the surroundings despite the sensing degradation, autonomous navigation then becomes possible. Model Predictive Control is used to enable precise trajectory tracking and robustness against disturbances. Path planning for collision-avoidance and autonomous exploration is facilitated through a receding horizon sampling-based algorithm that further accounts for the vehicle dynamic constraints and its limited endurance. The integrated system is tested in field experiments inside underground metal mines in Northern Nevada. The presented results correspond to the navigation and mapping of mine drifts and headings and involve environments that are completely dark, dust-filled, and particularly textureless. As shown, the proposed approach ensures the robust and reliable autonomous navigation of aerial robots in underground mines, alongside consistent mapping.