Learning Control-Affine Reduced-Order Models via Autoencoders
Researchers have developed a new framework for creating simplified control-affine reduced-order models (ROMs). This method employs autoencoders to map high-dimensional system states and inputs into a lower-dimensional latent space. The autoencoder and the state-space model are trained concurrently, with an extension to sequence-based modeling for enhanced prediction accuracy while maintaining the control-affine structure. The framework's effectiveness is demonstrated through feedback linearization and evaluated on numerical examples, comparing its prediction and control performance against a baseline. AI
IMPACT This research offers a novel approach to model reduction, potentially improving the efficiency of control system design and simulation for complex systems.