An integrated navigation system for articulated load–haul–dump machines in underground mines: adaptation and field validation

Abstract

Autonomous navigation of load–haul–dump (LHD) machines in underground mines is a key enabling technology for improving production efficiency, reducing personnel exposure to hazardous environments, and advancing intelligent mining systems. However, reliable autonomous operation remains challenging due to confined tunnel geometries, degraded perception conditions, and the complex kinematic characteristics of articulated vehicles. This paper presents a practical and fully integrated autonomous navigation system for articulated LHD machines operating in underground tunnels. A 3D LiDAR–based localization framework is developed using the FAST-LIO algorithm and extended with a global localization mode based on a prior point cloud map, enabling robust initialization and drift suppression in long, feature-poor roadways. To address the inefficiency and instability of probabilistic planners in narrow tunnels, a fast graph-search-based path planning method is proposed. By incorporating an euclidean distance transform map and a safe corridor generation strategy, the planner produces centered, collision-free paths that are further smoothed using cubic B-spline fitting under vehicle kinematic constraints. For motion execution, a Pure Pursuit–based tracking controller tailored to articulated LHD kinematics is employed. The proposed system is validated through both Gazebo simulations and real-world experiments conducted on an electric articulated LHD in a realistic underground tunnel environment. Experimental results show that the planning algorithm achieves computation times below 50 ms, while the path tracking error is maintained within 0.3 m under both forward and reverse driving conditions. These results demonstrate the feasibility, robustness, and industrial applicability of the proposed autonomous navigation framework for underground mining operations. Unlike existing underground LHD navigation approaches that rely on loop-closure-dependent simultaneous localization and mapping or stochastic planners with unpredictable computation time, the proposed framework uniquely integrates a prior-map-based global localization mode with a deterministic graph-search planning pipeline tailored to the confined geometry of linear mining tunnels.