Online System Identification
Overview
Online system identification studies how to infer and update a system’s dynamics while it is operating, so models and controllers can keep working as conditions change. Approaches include recursive and Bayesian estimation, neural state-space and Koopman models, Gaussian processes, and exploration strategies that choose informative actions while performing a task. These methods are used in settings such as robot and vehicle control, where model predictive control can use updated estimates to improve tracking and handle disturbances. A central tension is adapting quickly from limited data without destabilizing the system, overfitting noisy observations, or requiring unsafe excitation; reliable guarantees and computational efficiency remain important challenges.
Latest papers
23 papers3 papers in the last four weeks, level with the four weeks before. 0.0% of all new papers.