Linear learning control of robot motion (Q4037645)

The paper deals with trajectory following of a robot manipulator. Primary difficulty is that the robotic system is highly nonlinear and also presents inherently unknown dynamics, which is due to friction, backlash, and so on. This article uses Lyapunov's direct method together with Gronwall's inequality and derives a linear learning control law to achieve trajectory following for a robot manipulator. The control consists of two parts, a standard proportional-integral-differential (PID) control and a linear iterative term. By combining PID control and an iterative, learning control, it is shown that, under a reasonable condition, the position, velocity, and acceleration tracking errors can be made to approach zero asymptotically. It is shown that the proposed control is robust with respect to unknown nonlinear dynamics. The most important features of the developed learning control are that it is robust and easy to implement, does not require linear approximation or repeatability, and blends in with the classical PID controller. The paper concludes with a simulation study with a two degree-of-freedom manipulator. The simulation demonstrates the effectiveness of the proposed learning scheme. Although the paper offers a thorough theoretical study of the approach, the authors could also have paid attention to the well-known fact that \(D\) part in the PID control in practice is not implemented as a pure derivative of the error, but like a lead circuit [\textit{K. J. Åström} and \textit{B. Wittenmark}: Computer-controlled systems (1984)]. This would alter the authors proofs significantly.