From Batch-to-Batch to Online Learning Control: Experimental Motion Control Case Study

摘要

Data-driven feedforward control can significantly improve the positioning performance of motion systems. The aim of this paper is to exploit the concept of batch-to-batch learning control with basis function, applied in an online fashion. This enables learning within a task while maintaining task flexibility. A recursive least squares optimization is proposed on the basis of input/output data to compute the optimal feedforward parameters. The proposed method is successfully validated in simulation, and applied to a benchmark motion system leading to a major performance improvement compared to only feedback control.