Remarks on a Quaternion Recurrent Neural Network for Controlling a Robot Manipulator
Kazuhiko Takahashi, Risa Watanabe, Hidemi Yamasaki, Masafumi Hashimoto · 2019
In this study, the application of a quaternion recur-rent neural network (QRNN) for controlling a robot manipulator is investigated. A QRNN that considers two types of recurrent networks: Elman network and Jordan network, with a split-type quaternion activation function of neurons is used. Also, a realtime recurrent learning extended to quaternion numbers is introduced to train the network using a back-propagation algorithm. In the computational experiments, a three-link robot manipulator controlled using the proposed QRNN-based controller is used. The simulation results demonstrate the effectiveness of the proposed QRNN for practical control applications.