Learning Non-linear Dynamical Systems From Raw Images
Keywords:
machine learning, autoencoder, neural networks, latent space,non-linear systems, prediction, dynamical systemsAbstract
We introduce a method for model learning and control of non-linear dynamical systems from raw pixel
images. It consists of a deep generative model, belonging to the family of variational autoencoders, that learns to
generate image trajectories from a latent space in which the dynamics is constrained to be locally linear. Our model is
derived directly from an optimal control formulation in latent space, supports long-term prediction of image sequences
and exhibits strong performance on a variety of complex control problems.For capturing the information of non-linear
object’s behavior, we need to use high-dimensional data. Processing the high-dimensional data is expensive and not
feasible. So, in this model, first Auto-encoder is used for dimensionality reduction, and after prediction method
(transition mapping) is used, and the imagereconstructed. We demonstrate that our model enables learning good
predictive models of dynamical systems from pixel information only.
