不确定关节机器人模型的神经网络补偿自适应控制
Adaptively Controlling Neural Network Compensation with Uncertain Joint Robot Model

为了达到关节机器人轨迹跟踪控制的目的，针对由于机器人结构参数、作业环境干扰及结构共振模式等不确定性因素造成的机器人不确定性动力学模型，将该模型分解为名义模型和建模误差两部分，其中的建模误差采用RBF神经网络进行补偿和估计，得到其估计信息。RBF神经网络的权值通过Lyapunov稳定性分析和自适应算法进行调节。机器人的神经网络补偿自适应控制解决了机器人这类不确定模型的轨迹跟踪控制问题。对3关节机器人实验验证结果表明，3关节均在约4 s时跟踪期望轨迹，并且跟踪误差渐近趋近于0，并且RBF神经网络能很好地逼近由不确定性因素引起的建模部分。
In order to achieve the trajectory tracking control of a joint robot, because an uncertain joint robot's structural parameters cause a dynamic model's modeling errors and interfere with the working environment and the uncertain joint robot's resonant mode, the joint robot's dynamic model was divided into nominal model and error model. The error model was compensated by the RBF neural network, thus obtaining its estimation information. The neural network's output weights were adjusted adaptively according to the Lyapunov stability theory. The joint robot's adaptive neural network controller was used to solve the problems for the uncertain joint robot's dynamic system. Besides, the controller can gradually and stably track the desired trajectory though defining the Lyapunov function, being used to control a three-joint robot's torque. All the three joints can track the desired trajectory in 4 s. Tracking errors can gradually approach 0. Simulation and experimental results show that the RBF neural network can favorably approach modeling errors caused by uncertainties