Collision-Free Autonomous Navigation Solution for Mobile Wheeled

Abstract

Unmanned Ground Vehicles (UGVs) have become an imperative tool that is employed in a wide range of industrial sectors including oil and gas, agriculture, and defence. Their autonomy makes them an excellent choice for remote operations, particularly in complex outdoor environments. Therefore, the development of outdoor navigation solutions for UGVs has received the attention of many researchers in the field. This thesis is concerned with improving the outdoor localization of autonomous vehicles that use low-cost Global Positioning System (GPS) and Inertial Navigation System (INS) sensors in their operations. We propose a novel Kalman Filter (KF) based sensor fusion algorithm for low-cost and loosely coupled GPS/INS integration that tackles the linearization issues of a conventional Extended Kalman Filter (EKF) as well as addresses the errors associated with the Unscented Transformation (UT) of quaternion states in the traditional Unscented Kalman Filter (UKF) found in literature. Our algorithm termed the Augmented Quaternion Unscented Kalman Filter (AQUKF) offers an improved sensor fusion algorithm that accurately estimates both quaternion and non-quaternion states. Additionally, we consider the use of a multi-input Fuzzy Inference System (FIS) to recursively update the measurement noise covariance of our stochastic filter to match the noise statistics of the actual system. The resulting Fuzzy Adaptive Augmented Quaternion Unscented Kalman Filter (FA-AQUKF) reduces estimation uncertainties with the additional adaptive component, thus improving the accuracy of our localization solution. Initially, the performance of our proposed algorithms is evaluated using experimentally generated vehicle trajectories and validated against commercial solutions. Upon achieving satisfactory results, the algorithms are then implemented in real-time for the autonomous navigation of a Robot Operating System (ROS) operated UGV in the presence of static and dynamic obstacles. Results of all conducted experiments prove that our proposed algorithms deliver a significant improvement in vehicle state estimation and outdoor localization, besides satisfying the practical level of safety and accuracy desired for practical autonomous navigation applications.