Electrical Engineering and Systems Science > Systems and Control
[Submitted on 2 Apr 2020 (v1), revised 30 Aug 2021 (this version, v3), latest version 21 Aug 2023 (v8)]
Title:Safe Zero-Shot Model-Based Learning and Control: A Wasserstein Distributionally Robust Approach
View PDFAbstract:This paper presents a novel application of Wasserstein ambiguity sets to robustify online zero-shot learning and control. We identify and focus on scenarios of learning and controlling a system from scratch, starting with a randomly initialized model based on the strongest possible limitations on our prior knowledge of the dynamics. This paper labels this scenario as a "zero-shot" control problem, based on popular zero-shot transfer problems in machine learning. In this case, we adopt a loosely similar nomenclature to refer to a controller that must safely control a system it has truly never experienced or interacted with. Popular and current state-of-the-art methods in learning and control typically place more emphasis on model adaptation, and frequently require significant a-priori assumptions on knowledge of system dynamics and safe reference trajectories. Episodic designs are also commonplace in such applications, where constraint violation frequently occurs with gradually reduced frequency over the course of many sequential episodes of learning. We address the identified problem of single-episode zero-shot control by presenting a Wasserstein distributionally robust approach which, coupled with a receding horizon control scheme, can safely learn and control a dynamical system in a single episode.
Submission history
From: Aaron Kandel [view email][v1] Thu, 2 Apr 2020 00:39:07 UTC (195 KB)
[v2] Thu, 19 Aug 2021 01:03:00 UTC (1,578 KB)
[v3] Mon, 30 Aug 2021 21:30:47 UTC (1,578 KB)
[v4] Mon, 4 Jul 2022 23:16:04 UTC (1,775 KB)
[v5] Wed, 1 Feb 2023 21:03:20 UTC (2,004 KB)
[v6] Mon, 5 Jun 2023 21:31:11 UTC (2,005 KB)
[v7] Thu, 27 Jul 2023 21:40:26 UTC (2,009 KB)
[v8] Mon, 21 Aug 2023 18:01:05 UTC (2,280 KB)
Current browse context:
eess.SY
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.