SLSpy: Python-Based System-Level Controller Synthesis Framework

Shih-Hao Tseng,Jing Shuang Li

DOI: https://doi.org/10.48550/arXiv.2004.12565

2020-09-29

Abstract:Synthesizing controllers for large, complex, and distributed systems is a challenging task. Numerous proposed methods exist in the literature, but it is difficult for practitioners to apply them -- most proposed synthesis methods lack ready-to-use software implementations, and existing proprietary components are too rigid to extend to general systems. To address this gap, we develop SLSpy, a framework for controller synthesis, comparison, and testing. SLSpy implements a highly extensible software framework which provides two essential workflows: synthesis and simulation. The workflows are built from five conceptual components that can be customized to implement a wide variety of synthesis algorithms and disturbance tests. SLSpy comes pre-equipped with a workflow for System Level Synthesis (SLS), which enables users to easily and freely specify desired design objectives and constraints. We demonstrate the effectiveness of SLSpy using two examples that have been described in the literature but do not have ready-to-use implementations. We open-source SLSpy to facilitate future controller synthesis research and practical usage.

Systems and Control,Optimization and Control

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: Controller synthesis for large, complex, and distributed systems is a challenging task. Although many methods have been proposed in the literature, these methods often lack off - the - shelf software implementations, or the existing proprietary components are too rigid to be extended to general systems. This has led to a gap between theoretical research and practical applications, making it difficult for engineers and control practitioners to test, customize, or compare these methods.

To bridge this gap, the authors have developed the SLSpy framework, which aims to provide a highly extensible software framework for controller synthesis, comparison, and testing. SLSpy implements two main workflows: synthesis and simulation. These two workflows are constructed from five conceptual components and can be customized to implement various synthesis algorithms and interference tests. SLSpy also comes pre - installed with a workflow based on system - level synthesis (SLS), enabling users to easily and freely specify design goals and constraints.

Through two examples, the authors demonstrate the effectiveness of SLSpy. These examples are described in the literature but do not have off - the - shelf implementations. In addition, the open - source nature of SLSpy is conducive to future controller synthesis research and practical applications.