Motion controls using complex-valued neural networks with feedback loops

A smooth motion control method is proposed using complex-valued layered neural networks that have feedback loops. The complex-valued neural networks express the continuous nature of physical fields and motions. A continuous backpropagation learning process is applied for the network with a set of time-sequential input and output data required to express intended motions. During the steps of the learning process, the values of error functions for time-sequential output vectors reduce almost monotonically, resulting in successful learning. The learning process of sinusoidal trajectories with various frequencies is demonstrated