The vast and difficult topography of mining sites have rendered conventional automation solutions such as conveyor belts and stationary robotic systems impractical; highlighting mobile robots as the optimal solution. This technical project, propose an autonomous mobile robotic model for mapping and navigation in mining sites using machine learning. This is achieved in real time by deploying and inferencing artificial neural networks. The inference is done on Nvidia Jetson platform for AI-based embedded applications. The neural network is based on You-Only-Look-Once real-time object detection system and trained with a custom-made dataset. The dataset consists of images for trucks, loaders, mining drills, crushers and excavators taken from local mining sites in Oman and sources like Google Open Images Dataset. The proposed model runs on Robot Operating System and incorporates sensor fusion by integrating stereo imaging, lidar and odometry for environment perception, 2D and 3D mapping, path planning and navigation. The proposed model was thoroughly evaluated on three different tests comprising several hours of mapping and navigation. The experiment show that the presented approach significantly outperforms conventional automation solutions such as conveyor belts and stationary robotic systems in terms of cost, uptime and productivity.