Cascaded ILC-MPC controller for robot manipulators

Trajectory tracking is one of the key aspects of robot control system design. High-precision trajectory under disturbances, unmodeled dynamics, and parametric uncertainties of robotic systems is a highly challenging task for multiple DoFs. Many different control approaches, from simple PID controllers to advanced controllers such as Sliding Mode Control, Adaptive Control, Robust Control, Fuzzy Control, and Model Predictive Control (MPC), are used within trajectory tracking problems for robotic systems [1]. One of the control strategies suitable for trajectory tracking problems for systems executing repetitive tasks is Iterative Learning Control (ILC) [2]. The ILC is an intelligent, memory-based control method aimed at improving the transient response performance of the system that operates repetitively over a fixed time interval [3]. Several papers tackle improving the MPC with iterative learning controllers [4], [5]. This research presents the integration of MPC and ILC using a simple structure consisting of a cascaded ILC and MPC controller (Fig. 1.a) for position control of the robot manipulator. The motivation for the cascaded structure is that the ILC controller does not interfere with the MPC controller’s nonlinear optimization solver since the parallel structure of the ILC and MPC would make MPC’s objective function more complex. The proposed control system is verified via numerical simulation on a two-degree-offreedom planar manipulator (Fig. 1.b). The constant disturbance (time and iteration-wise) is considered in the simulation model. The reference trajectory in joint space is set to be a cubic polynomial. The performance of the proposed control system is evaluated using Maximal Absolute Error (MAE). For each iteration, MAE is calculated and plotted as shown in Fig. 2.a, from which it can be concluded that the ILC controller reduces the tracking error for both joints in approximately four iterations. Fig. 2.b shows that the first joint tracks the reference trajectory almost perfectly while the second joint has a larger tracking error, although still significantly reduced in comparison to the sole MPC controller.The proposed Cascaded ILC-MPC control solution for robotic systems shows promising results. Further investigation should encompass the reduced dynamics model in the MPC and the fractional derivative in the ILC.