基于在线学习的柔性关节机器人自适应神经轨迹跟踪控制

Adaptive neural trajectory tracking control of flexible joint robot based on online learning

针对传统机器人关节控制算法跟踪精度低、鲁棒性差的缺点,基于自适应神经网络提出了一种机器人关节轨迹跟踪算法。算法由自适应神经网络和在线学习反馈模块组成,自适应神经网络将自适应函数同神经网络结合,提高了神经网络训练准确性。同时通过在线学习反馈模块实时更新非线性基函数的内部权值,以进一步减小跟踪误差。再采用时间尺度分离减少了神经网络和在线学习的耦合误差,使得内部权重低于输出层权重的更新速度,从而使模型结构能够迅速适应未知的动态变化与干扰。仿真实验表明,所提算法与对比算法相比误差值要低约60%,说明了该算法可以提高机器人轨迹跟踪精度,降低误差。

Aiming at the shortcomings of low tracking accuracy and poor robustness of traditional robot joint control algorithm,a robot joint trajectory tracking algorithm based on adaptive neural network is proposed.The algorithm is composed of adaptive neural network and online learning feedback module.Adaptive neural network combines adaptive function with neural network to improve the accuracy of neural network training.At the same time,the online learning feedback module updates the internal weights of the nonlinear basis function in real time to further reduce the tracking error.Then time scale separation is used to reduce the coupling error between neural network and online learning,so that the internal weight is lower than the update speed of the output layer weight,so that the model structure can quickly adapt to unknown dynamic changes and disturbances.Simulation results show that the error propagation value of the algorithm is about 60%lower than that of the comparison algorithm,which shows that the algorithm can improve the trajectory tracking accuracy of the robot and reduce the error.