Addressing the challenge of autonomous navigation for mining Unmanned Ground Vehicles (UGVs) in mining environments, this paper introduces a comprehensive navigation framework tailored to bumpy terrains, integrating improved obstacle recognition with localized obstacle avoidance techniques. The algorithmic realization of this framework involves four key stages. The first stage involves 3D mapping and localization of the environment. It is achieved through the fusion of a radar LiDAR-Visual Inertial SLAM System (LVI-SAM) and Monte Carlo Localization (MCL) algorithms. In the next stage, dense global paths are generated using Plane Fitting Rapidly Expanded Random Tree (PF-RRT*) and Gaussian Process Regression (GPR). In the third stage, an improved Density-Based Spatial Clustering of Applications with Noise (DBSCAN*) is employed to address the obstacles around the UGV, and these obstacles are encapsulated using adaptive cylindrical envelopment. In the final stage, an Adaptive Control Barrier Function (A-CBF) is proposed to be used in combination with Model Predictive Control (MPC) to aid in dynamic obstacle avoidance and global trajectory tracking. The algorithmic framework’s effectiveness is confirmed through simulation experiments in the Gazebo software and real-world navigation tests conducted in mining environments, with the results exhibiting the UGV’s precise target-reaching abilities.