【摘要】 为了实现对不同结构的两自由度机械臂的控制,基于机器视觉技术和强化学习技术设计了智能控制器。该智能控制器的强化学习模型是深度确定性策略梯度算法,使用卷积神经网络设计了的传统深度确定性策略梯度算法的演员-评论家网络结构,以此使得深度确定性策略梯度算法能够处理图像矩阵信息。以两自由度机械臂的控制理论为前提,利用不同结构的两自由度机械臂控制输入值数量相同的特点,并结合设计了的深度确定性策略梯度算法设计出适用于不同结构的两自由度机械臂智能控制器。仿真结果表明了该智能控制器在不配置任何两自由度机械臂结构模型先验知识的情况下,分别就三种不同结构两自由度机械臂进行训练,训练完成后该智能控制器确实能够控制三种不同结构的两自由度机械臂并使其迅速到达目标位置,验证了使用同一个智能控制器经过不同的训练后能够实现对不同结构的两自由度机械臂的控制。更多还原

【Abstract】 In order to control the two DOF manipulator with different structures, an intelligent controller was designed based on machine vision and reinforcement learning technology. The reinforcement learning model of the intelligent controller is the Deep Deterministic Policy Gradient algorithm. Convolutional neural network is used to reconstruct the actor-critic network structure of the traditional Deep Deterministic Policy Gradient algorithm, so that the Deep Deterministic Policy Gradient algorithm can process the image matrix information. Based on the control theory of the two-degree-of-freedom manipulator, the intelligent controller of the two-degree-of-freedom manipulator with different structures is designed by using the characteristics of the same number of control input values of the two-degree-of-freedom manipulator with different structures and combining with the reconstructed depth deterministic strategy gradient algorithm. Simulation results show that the intelligent controller in not configure any two degrees of freedom mechanical arm structure model under the condition of prior knowledge, then three different structure respectively two degrees of freedom mechanical arm training, training after the completion of the intelligent controller can control three different structures of two degrees of freedom mechanical arm and make it quickly reach the target location, use the same intelligent controller is verified through different training can achieve control of the different structure of two degrees of freedom mechanical arm.更多还原