Remarks on Robot Controller Application of Clifford Multi-layer Neural Networks

In this paper, Clifford multi-layer neural networks using back-propagation algorithm are applied to inverse kinematics control of a robot manipulator as a first step of utilizing Clifford neural networks for robot control applications. The control system based on the on-line specialized learning architectures is considered and its characteristics are investigated. To increase the success rate of learning in this architecture, the weight-resetting methods are introduced into the drawback learning of the Clifford multi-layer neural network. In the computational experiments, the training of Clifford neural networks converges with a fewer number of iterations compared with the real number neural network which has more complex network topology and more parameters. The Clifford algebra framework makes the neural computing more efficient in inverse kinematics controller which shows the potential of the Clifford multi-layer neural network in robot control system.