Learning Control-Affine Reduced-Order Models via Autoencoders

2026-06-03 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, medium

Summary

Ali Mjalled and Martin Mönnigmann introduce a framework for identifying control-affine reduced-order models (ROMs) using autoencoders (AEs). This method transforms high-dimensional states and inputs into reduced latent representations, suitable for control-affine state-space dynamics, through the simultaneous training of the AE and the state-space model. The framework extends discrete ROM formulations to sequence-based models, which process state and input histories to enhance prediction accuracy while preserving the control-affine structure. The authors motivate their approach by applying feedback linearization to the derived models and provide guidelines for efficient implementation. The framework's performance is evaluated on two numerical examples, comparing its prediction accuracy and control effectiveness against a baseline model that uses an AE to identify a latent space with linear state-space dynamics.

Key takeaway

For Machine Learning Engineers developing reduced-order models for complex control systems, this framework offers a robust approach to maintain control-affine properties. You should consider simultaneously training autoencoders with state-space models to efficiently reduce dimensionality while preserving dynamic structure. Implementing sequence-based models can further enhance your prediction accuracy, making your ROMs more effective for feedback linearization and system control.

Key insights

Simultaneous training of autoencoders and state-space models enables robust control-affine reduced-order model identification.

Principles

• Control-affine structure is preserved in reduced-order models.
• Sequence-based models improve prediction accuracy.
• Feedback linearization can be applied to derived models.

Method

The method involves simultaneously training an autoencoder to reduce high-dimensional states/inputs and a state-space model to learn control-affine dynamics in the latent space, extending to sequence-based models.

In practice

• Apply feedback linearization to derived ROMs.
• Utilize sequence-based models for enhanced prediction.

Topics

• Reduced-Order Models
• Autoencoders
• Control-Affine Systems
• State-Space Models
• Feedback Linearization
• Machine Learning Control

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Robotics Engineer

Related on AIssential

• Robustness of Agentic AI Systems via Adversarially-Aligned Jacobian Regularization — AAJR enhances agentic AI robustness by controlling sensitivity only along adversarial ascent directions, improving stability and performance.
• Autoencoders - Explained — Autoencoders learn data structure by compressing and reconstructing inputs, organizing data by meaning in a latent space.
• A Nonlinear Separation Principle: Applications to Neural Networks, Control and Learning — A nonlinear separation principle enhances stability and performance in recurrent neural networks for control and learning.
• Koopman Operator Identification of Model Parameter Trajectories for Temporal Domain Generalization (KOMET) — KOMET enables zero-retraining model adaptation by predicting parameter trajectories via Koopman operator identification.
• Closed-form predictive coding via hierarchical Gaussian filters — Closed-form predictive coding via hierarchical Gaussian filters enables deep, efficient, and biologically plausible neural network training.
• Multi-Agent Model-Based Reinforcement Learning with Joint State-Action Learned Embeddings — MMSA combines model-based MARL with joint state-action representation learning for enhanced coordination and sample efficiency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.