Deep Learning for Posture Control Nonlinear Model System and Noise Identification

Vittorio Lippi, Thomas Mergner, Christoph Maurer


In this work we present a system identification procedure based on Convolutional Neural Networks (CNN) for human posture control models. A usual approach to the study of human posture control consists in the identification of parameters for a control system. In this context, linear models are particularly popular due to the relative simplicity in identifying the required parameters and to analyze the results. Nonlinear models, conversely, are required to predict the real behavior exhibited by human subjects and hence it is desirable to use them in posture control analysis. The use of CNN aims to overcome the heavy computational requirement for the identification of nonlinear models, in order to make the analysis of experimental data less time consuming and, in perspective, to make such analysis feasible in the context of clinical tests. After testing the performance of the CNN on validation and test sets, two examples are presented and discussed from the qualitative point of view: the identification of parameters using data from human experiments and using data of a simulated model with some differences with respect to the one used to build the training set. Some potential implications of the method for humanoid robotics are also discussed.


Paper Citation