Determination of Elastodynamic Model Parameters using a Recurrent Neuro-Fuzzy System
Andreas Nürnberger, Arne Radetzky · 1999
Elastodynamic models are used for real-time simulation of deformable objects in virtual reality applications. To obtain a realistic simulation, the physical parameters of the model must be defined appropriately. Furthermore, the usability of elastodynamic models in virtual reality applications depends on the used simulation algorithm to a great part, since interactions with the simulated object have to be done in real-time. In this paper we present a learning algorithm which determines the parameters of a spring-mass model. The algorithm is based on a recurrent neural network and can be initialized by use of a fuzzy system. The presented algorithm uses the positions of specific object points during discrete time steps to learn the required parameters. The time series data for learning can be derived, for example, by a time dependent optical measurement of an object under influence of external forces.