Tube MPC for Bilinear Koopman Models using Robust Control Contraction Metrics

What the math gives to ML

The paper's transferable asset is a constructive contraction-metric framework for learned nonlinear dynamics: instead of merely fitting trajectories, it learns a state-dependent metric and feedback law that quantify how model mismatch propagates through the system. This suggests adding a contraction certificate to latent world models or recurrent predictors, so perturbations, rollout errors, and nearby latent trajectories are actively shrunk rather than amplified. The geodesic computation is a practical numerical subroutine: represent paths in a polynomial basis and minimize their metric energy under endpoint constraints. The most credible first transfer is a contraction-regularized latent dynamics model, evaluated by long-horizon rollout stability and robustness to injected state or model noise.