Multifidelity Robust Controller Design with Gradient Sampling

Robust controllers that stabilize dynamical systems even under disturbances and noise are often formulated as solutions of nonsmooth, nonconvex optimization problems. While methods such gradient sampling can handle the nonconvexity nonsmoothness, costs evaluating objective function may be substantial, making robust control challenging for with high-dimensional state spaces. In this work, we introduce multi-fidelity variants leverage low-cost, low-fidelity models low-dimensional spaces speeding up process while nonetheless providing convergence guarantees a high-fidelity model system interest, which is primarily accessed in last phase process. Our first method initiates on higher fidelity starting points obtained from cheaper, lower models. second relies ensembles gradients computed low- Numerical experiments controlling cooling steel rail profile laminar flow cylinder wake demonstrate our new achieve to two orders magnitude speedup compared single-fidelity alone.