MetaKoopman: Bayesian Meta-Learning of Koopman Operators for Modeling Structured Dynamics under Distribution Shifts

Researchers introduced MetaKoopman, a Bayesian meta-learning framework that learns a prior over Koopman operators for nonlinear dynamics.

The model adapts using recent trajectories and outputs predictive uncertainty, improving both forecasting and calibration.

In truck-and-trailer experiments and simulated control tasks with snow, ice, and other shifts, it outperformed prior methods.

The approach could make autonomous control and motion planning safer and more reliable in changing real-world conditions.