On learning transient and cutoff frequency tuning in ILC

Many iterative learning control (ILC) design methods employ a low-pass filter to guarantee good learning behavior. To determine the cutoff frequency of the filter, a system model is often required. Because the inaccuracy of system model in practice, the learnable bandwidth obtained from the model is often chosen as a conservative value. This can severely underachieve in learning performance. In this paper, we address the learning transient and cutoff frequency tuning. Based on this, the learnable bandwidth of system can be tuned automatically so that the balance can be reached between good learning transient and tracking accuracy. The attractive properties include that the system model is not necessary and that the tracking error during this process is kept in an acceptable level. Experiments on a SCARA robot are carried out to illustrate the tuning process.