Distributionally Robust Chance Constrained Data-Enabled Predictive Control

In this article we study the problem of finite-time constrained optimal control unknown stochastic linear time-invariant (LTI) systems, which is key ingredient a predictive algorithm—albeit typically having access to model. We propose novel distributionally robust data-enabled (DeePC) algorithm uses noise-corrupted input/output data predict future trajectories and compute inputs while satisfying output chance constraints. The based on 1) nonparametric representation subspace spanning system behavior, where past are sorted in Page or Hankel matrices; 2) optimization formulation gives rise strong probabilistic performance guarantees. show that for certain objective functions, DeePC exhibits out-of-sample performance, at same time respects constraints with high probability. provides an end-to-end approach design LTI systems. illustrate closed-loop aerial robotics case study.