Model-based Feedforward Control in Industrial Robotics

Abstract

Simple linear joint controllers are still used in typical industrial robotic systems. The use of these controllers leads to non-negligible dynamic path deviations for applications that require high path accuracy. These deviations result from the strong influence of nonlinearities, such as multi-body dynamics and gear friction. Sophisticated nonlinear control algorithms, known from the literature, are still not used because they usually require an expensive change of the control architecture. Therefore, different compensation methods are compared in this paper which reduce the path deviations by correction of the desired trajectory. This means that no torque interface is required, only an interface for path corrections is necessary. Such an interface normally exists so that the methods can simply be implemented within standard industrial controls. Using the industrial robot Siemens manutec-r15 the methods are experimentally compared with respect to their efficiency and practical applicability. Starting from this, one method is chosen for application to the stateof-the-art industrial robot KUKA KR15. The algorithm is based on a complete nonlinear dynamic model of the robot. It is integrated into the standard control KRC1. The experimental results prove the efficiency and the industrial applicability of the method.

Assembling wheels to continuously conveyed car bodies using a standard industrial robot

“…Again, a feed-forward controller is used [19], [20]. In contrast to the internal joint controllers which at time step k process the joint positions q c , the use of desired joint trajectories allows to restrict to feed-forward.…”

Spatial Vision-Based Control of High-Speed Robot Arms

“…A standard sensor interface as (Pertin & Bonnet des Tuves, 2004) q . For curved paths to be executed at high speed we recommend to increase accuracy using advanced control methods which process not only the current desired pose but also the desired speed and acceleration or -what we use -the desired poses of multiple future sampling steps as (Clarke et al, 1987) or (Grotjahn & Heimann, 2002). Fig.…”

Is Impedance-Based Control Suitable for Trajectory Smoothing?

“…q c is the vector of commanded positions which is the usual input to industrial control systems similar to that of Pertin and Bonnet-des-Tuves (2004). q c is computed using an adaptive feedforward controller presented in Lange and Hirzinger (1996) or Grotjahn and Heimann (2002), which processes in each sampling step the desired values of the current and of some future sampling steps (bold face lines in Figure 1 denote values of several sampling steps). A sufficient number of future values theoretically allows the realization of an ideal robot, i.e.…”