Design and practical implementation of a Neural Network self-tuned Inverse Dynamic Controller for a 3-DoF Delta parallel robot based on Arc Length Function for smooth trajectory tracking

This paper proposes an online Neural Network self-tuned Inverse Dynamic Controller (IDC) for high-speed and smooth trajectory tracking control of a 3-DoF Delta robot. The foregoing approaches provides a suitable controller for a wide range of nonlinear paths and reduce the end-effector oscillations at high speed. To this end, a compact and accurate dynamic model of the system is derived by taking into account actuators and gearbox dynamics. In order to alleviate some drawbacks of a velocity-based controller, such as not being able to track highly dynamic paths, an Inverse Dynamic Controller (IDC) is designed which can perform fast maneuvers accurately. The proposed IDC controller is practically implemented on the robot in following nonlinear paths comparing to the velocity-based controller. Afterward, controller parameters are tuned by resorting to the so-called Arc Length Function (ALF) in order to improve the smoothness of tracking the prescribed path. After that, a Feedforward Neural Network (NN) is trained with the help of the system's model and Arc Length Function (ALF) to adjust controller coefficients in real-time implementation adaptively. By comparing the Root Mean Square Error (RMSE) results, it can be inferred that the proposed methods can reduce the end-effector oscillations up to 60 percent in practical implementation compared to other dynamic and kinematic methods. As a result, RMSE error is reduced from 0.00603 for the kinematic controller to 0.00063 by applying the NN-IDC.