كنترل ايمن بازوي ربات با استفاده از يادگيري تقويتي عميق

This research focuses on the development an‎d implementation of an intelligent control framework for the safe control of robotic arms in dynamic environments using deep reinforcement learning. Given the pivotal role of robotic arms in modern industries, ensuring their safety during interaction with the environment an‎d moving obstacles remains a fundamental challenge for this technology. In this study, to find the optimal control policy, advanced reinforcement learning algorithms, including DDPG, TD3, an‎d SAC, were initially eva‎luated. The results of this assessment demonstrated the significant superiority of the SAC algorithm in addressing this problem; therefore, SAC was selec‎ted for training the control policy of the Franka Pan‎da robotic arm. The main innovation of this research is the design of an integrated control architecture where a state observer estimates joint velocities online, an‎d this information is directly incorporated into the reward function to penalize abrupt an‎d risky movements. This approach guides the robot toward learning intrinsically safe an‎d smooth motions. The systemʹs performance was eva‎luated in the PyBullet simulation environment within complex scenarios involving reaching a target an‎d obstacle avoidance. Simulation results indicate that the proposed method, based on the SAC algorithm, successfully accomplishes its task with a high success rate an‎d generates significantly safer an‎d smoother trajectories compared to other algorithms. Furthermore, the practical feasibility of this approach was confirmed through the implementation an‎d validation of its key components on real hardware. This achievement contributes to the development of a new generation of intelligent an‎d reliable robots, marking an effective step towards realizing the goals of Industry 5.0, namely safe an‎d efficient human-robot collaboration.

5