Browsing by Author "Soliman, Ahmed Adel Ahmed Fahmy"
Now showing 1 - 1 of 1
- Results Per Page
- Sort Options
PhD DissertationPublication Metadata only Optimal trajectory generation and adaptive control of an underactuated and self-balancing lower body exoskeletonSoliman, Ahmed Adel Ahmed Fahmy; Uğurlu, Regaip Barkan; Uğurlu, Regaip Barkan; Bebek, Özkan; Ünal, Ramazan; Erbatur, K.; Şafak, K. K.; Department of Mechanical EngineeringThis thesis presents an approach for developing three-dimensional (3D) dynamic walking capabilities in a bipedal exoskeleton with underactuated legs. The proposed framework consists of a trajectory generator and an optimized inverse kinematics algorithm designed to handle underactuation. To achieve feasible task velocities despite the underactuated legs, the inverse kinematics algorithm utilizes a task prioritization method by exploiting the null space. This approach allows lower-priority tasks, such as swing foot orientation, to be accomplished to the greatest extent possible without interfering with higher-priority tasks like the Center of Mass trajectory. Simultaneously, the trajectory generator analytically incorporates the zero moment point concept, ensuring continuous acceleration throughout the entire walking period, regardless of changes in contact and phase. Furthermore, three locomotion controllers were developed to complement the proposed task prioritization algorithm and enhance its robustness against significant parameter uncertainty and external disturbances. These controllers include the zero moment point impedance feedback controller, along with two other state-of-the-art locomotion controllers: admittance control and centroidal momentum control. The objective is to integrate these controllers with the task prioritization algorithm, collectively improving the system's ability to handle challenging conditions and uncertainties. A series of simulation experiments were conducted using a 3D simulator to verify the validity and robustness of these controllers for thorough benchmarking. A human-robot coupled model is considered, including a 40 kg underactuated exoskeleton and 12 distinct anthropomorphic subjects. When combined with the proposed task priority-based optimization algorithm, all three controllers demonstrate adequate performance in addressing balanced locomotion behavior. The proposed zero moment point impedance controller shows statistically significant results, indicating a comparatively more robust feature. As the proposed locomotion controller approaches are model-based controllers, there is a desire to develop a real-time applicable inertial parameter identification algorithm to improve the locomotion controller's adaptability against inertial parameter variations. A semidefinite programming algorithm is developed to perform the identification algorithm recursively while guaranteeing the complete physical consistency of the identified inertial parameters. A recursive algorithm to update the identifiability projection matrix is developed to manage the inclusion of newly acquired samples and arrange them concerning the old identifiable parameters. The idea of the filtered regressor is used to mitigate the effect of contact transitions and noise without losing information about the identifiable parameters. To verify the validity of the proposed identification algorithm, a series of simulation experiments are conducted using 3D simulator using the human-robot coupled model. As a result, the algorithm shows feasible performance based on accuracy, computation time, and the complete physical consistency of the identified parameters. To verify the proposed algorithm, an exoskeleton prototype was constructed. The prototype was equipped with eight series elastic actuators, sixteen force-sensitive resistors for measuring contact force, and another sixteen absolute encoders for measuring motor angular displacements and series elastic actuator spring deflections. Communication with series elastic actuators and sensors was enabled through a set of interface circuits and a desktop PC. The real-time operation was justified by employing Ubuntu 18.04 and Xenomai 3.1. The required Cartesian and joint-level controllers were programmed using the C language with the GNU scientific library. Simultaneous resolution of both algorithms was facilitated through parallel programming. Experimental development of squat, sway, and sagittal walk motions was carried out. The real-time applicability and feasibility of the proposed algorithms, as well as the developed hardware, were demonstrated by the experiments.