Infinite-Horizon Distributionally Robust Regret-Optimal Control

Authors: Taylan Kargin, Joudi Hajar, Vikrant Malik, Babak Hassibi

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we present the performance of the DRRO controller, compared to H2, H , regret-optimal and other finite-horizon DR controllers. We present frequency domain and time-domain evaluations, and we showcase the performance of the rational approximation method. We employ benchmark models such as [REA4], [AC15], and [HE3] from (Leibfritz & Lipinski, 2003).
Researcher Affiliation Academia Taylan Kargin * 1 Joudi Hajar * 1 Vikrant Malik * 1 Babak Hassibi 1 1California Institute of Technology. Correspondence to: Taylan Kargin <tkargin@caltech.edu>.
Pseudocode Yes We conceive Algorithm 1, a procedure based on the Frank-Wolfe method, to compute the optimal M in the frequency domain. Detailed pseudocode is provided in Algorithm 1 in Appendix F.1.
Open Source Code No The paper does not provide any concrete statement or link regarding the availability of its source code.
Open Datasets Yes We employ benchmark models such as [REA4], [AC15], and [HE3] from (Leibfritz & Lipinski, 2003).
Dataset Splits No The paper does not provide specific dataset split information (percentages, sample counts, or citations to predefined splits) for training, validation, or testing. The term “validation” is not used in the context of data splits.
Hardware Specification Yes We perform all experiments using MATLAB, on an Apple M1 processor with 8 GB of RAM.
Software Dependencies No “We perform all experiments using MATLAB”. This mentions software but lacks specific version numbers for MATLAB or any other dependencies.
Experiment Setup Yes We specify the nominal distribution as a Gaussian, with zero mean and identity covariance. The finite-horizon controllers operate over a horizon of only s = 30 steps and are re-applied every s steps. For the [AC15] system, the worst-case expected regret cost, as outlined in (2.1), for DR-RO, the H2, H , and RO controllers. are depicted in Figure 1b. We observe that for smaller r, the DR-RO performs close to the H2 controller.