Data-Driven Feedforward Optimization Algorithm for Industrial Robots

To solve the problem of insufficient joint tracking performance in the feedforward control of industrial robots caused by model and control quantity errors, a data-driven feedforward optimization algorithm is proposed. This algorithm not only enhances joint tracking performance but also optimizes trajectory accuracy. Firstly, aiming at the nonlinear residuals in the dynamic model resulting from linear assumptions, neglect of high-order terms and other factors, a joint torque residual fitting method based on ensemble learning is proposed. This method constructs a hybrid dynamic model that combines mechanism-based and data-driven approaches, thereby improving the accuracy of torque feedforward quantities. Secondly, based on the principle of iterative learning control, a data-driven velocity feedforward optimization algorithm is proposed. By comparing historical and real-time data, the learning gain is dynamically adjusted, and the velocity feedforward parameters are updated iteratively to reduce joint tracking deviations. Finally, experiments on industrial robot verify that the torque residual fitting based on ensemble learning reduces the joint torque prediction error by more than 77%. After optimizing the feedforward control quantities, the root mean square error of joint tracking is significantly improved: it is reduced by more than 60% for linear trajectories and by more than 20% for circular trajectories.