An hypercomplex neural network platform for robot positioning

In this paper the attitude control problem of a rigid body in 3-D space is approached by introducing a new neural tool (HMLP) developed in quaternion algebra. Such a choice allows one to deal efficiently with the attitude control problem, decreasing the computational complexity with respect to the rotation matrix representation. The proposed neural tool is based on a cascade of several quaternionic neural networks, representing both the system and the controller, where only the HMLP representing the controller has to be trained. The neural controller allows one to obtain the desired attitude of a rigid body, whose model is unknown, in a finite number of steps.