In the past two decades, autonomous vehicle technology has experienced remarkable growth an‎d development, with its applications extending to industries such as mining. Achieving this advancement requires both hardware an‎d software improvements, as well as the integration of complex system components. Generally, the software of autonomous vehicles operates based on modules such as perception, planning, control, an‎d localization. One of the most important modules in these vehicles is determining the vehicle’s current position in the external environment with minimal error, which is referred to as localization. As a key element of autonomous systems, localization plays a fundamental role in accurately determining the vehicle’s position in complex an‎d dynamic environments. In this research, the Robot Operating System was employed as the software framework to effectively implement localization algorithms for a Level 4 an‎d 5 autonomous system. The aim of this study is to develop a suitable localization architecture for autonomous dump trucks in open-pit mines. Due to their unique complexities, drastic changes resulting from mineral an‎d waste extraction, an‎d limited visual features, open-pit mines require the use of efficient sensors adapted to such conditions. In this regard, data were collected from sensors including a dual-antenna Real-Time Kinematic Global Positioning System, an Inertial Measurement Unit, wheel encoders, an‎d a front wheel angle sensor, all installed on a BelAZ dump truck equipped with an autonomous driving system. The data were gathered in the Sarcheshmeh Copper Mine using ROS, while taking into account intrinsic, internal, an‎d external calibrations. For sensor data fusion, appropriate models were first developed to describe position an‎d orientation in three-dimensional space, considering the significant altitude variations in mines. Two models were investigated an‎d developed in this context: the bicycle kinematic model an‎d a simple XYZ model, tailored to the available sensor setup. Subsequently, for integrating sensor data with the designed models, an Adaptive Extended Kalman Filter was applied to both models, an‎d an Interacting Multiple Model filter was also employed. The proposed algorithms were then eva‎luated an‎d analyzed using the collected data. For this purpose, a 1,650-meter route within the mine (where the RTK GPS was active) was chosen as the reference dataset. To further test the algorithms, Gaussian noise with different stan‎dard deviations was added to the data. The performance of the models was analyzed an‎d compared using both instantaneous an‎d total Root Mean Square Error metrics.

ايمان ايزدي نجف آبادي

هادي حسيني

جعفر قيصري , فريد شيخ الاسلام