كنترل هيبريد نيرو/موقعيت مبتني بر يادگيري تقويتي سيستم رباتيكي

This dissertation presents the design an‎d development of a novel intelligent control framework for hybrid force/position control of robotic systems, aiming to overcome the fundamental limitations of classical controllers, including reliance on precise models, sensitivity to uncertainties an‎d disturbances, an‎d the undesirable phenomenon of chattering. The central paradigm of this research lies in integrating the adaptability an‎d learning capabilities of reinforcement learning with the robustness of advanced sliding mode control. In this context, two distinct intelligent controllers are designed an‎d introduced: 1. A hybrid controller for a Delta parallel robot: This controller employs the Deep Deterministic Policy Gradient (DDPG) reinforcement learning algorithm to online optimize the parameters of a fixed-time sliding mode controller. The ultimate goal of this strategy is the simultaneous reduction of force an‎d position tracking errors, complete elimination of chattering, an‎d significant reduction of control effort, while guaranteeing robust system stability against parametric uncertainties an‎d external disturbances. 2.A semi model-free controller for a redundant serial robot (OpenManipulator-X): This novel controller combines super-twisting sliding mode control with reinforcement learning to optimally learn the null-space torque. Its most notable feature is the elimination of the explicit an‎d fragile separation of force an‎d position control subspaces. In addition to achieving accurate tracking, chattering elimination, an‎d reduced control effort, this controller is specifically designed to address time delays in actuation an‎d communication, thereby ensuring system stability under such conditions. Furthermore, by exploiting the null space, it enables optimization of secondary objectives such as energy efficiency (control effort minimization). The superior performance of the proposed controllers is validated through extensive simulations under various scenarios an‎d comprehensive comparisons with classical methods. The results clearly demonstrate that the proposed framework constitutes a significant step toward the development of self-adaptive, robust, an‎d efficient robotic systems capable of performing complex interactive tasks in real-world environments

6

153429