Dynamic modeling of robotic trajectories using the parametrized SOM

Abstract

Planning and control of robotic trajectories is an important and open issue. This paper uses an unsupervised neural network model to construct the dynamical modelling of trajectories. A neural network with a short term memory mechanism, was designed to provide the associated joint angles when it receives as input the present and some past states of the robot spatial position. The model uses the self-organizing map (SOM) to approximate the mapping using just some states of the trajectory.