Scholarlink

The action governorGovernor is an add-on scheme to a nominalNominal control loop that monitors and adjusts the control actions to enforce safetySafety specifications expressed as pointwise-in-time state and control constraints. In this chapter, we introduce theRobust Action Governor (RAG) RobustRobust Action GovernorGovernor (RAG) for systems the dynamics of which can be represented using discrete-time Piecewise Affine (PWA) modelsModels with both parametric and additive uncertainties and subject to non-convex constraints. We develop the theoretical properties and computational approaches for the RAG. After that, we introduce the use of the RAG for realizing safeSafe ReinforcementReinforcement LearningLearning (RL), i.e., ensuring all-time constraintConstraint satisfaction during online RL exploration-and-exploitation process. We illustrate the effectiveness of the RAG in constraintConstraint enforcement and safeSafe RL using the RAG by considering their applications to a soft-landing problem of a mass–spring–damper system.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Robust Action Governor for Uncertain Piecewise Affine Systems with Non-convex Constraints and Safe Reinforcement Learning

  • 章节

  • 发表日期：

    2025年

查看全文
  • Yutong Li,
  • Nan Li,
  • H. Eric Tseng,
  • Anouck Girard,
  • Dimitar Filev,
  • Ilya Kolmanovsky

摘要

The action governorGovernor is an add-on scheme to a nominalNominal control loop that monitors and adjusts the control actions to enforce safetySafety specifications expressed as pointwise-in-time state and control constraints. In this chapter, we introduce theRobust Action Governor (RAG) RobustRobust Action GovernorGovernor (RAG) for systems the dynamics of which can be represented using discrete-time Piecewise Affine (PWA) modelsModels with both parametric and additive uncertainties and subject to non-convex constraints. We develop the theoretical properties and computational approaches for the RAG. After that, we introduce the use of the RAG for realizing safeSafe ReinforcementReinforcement LearningLearning (RL), i.e., ensuring all-time constraintConstraint satisfaction during online RL exploration-and-exploitation process. We illustrate the effectiveness of the RAG in constraintConstraint enforcement and safeSafe RL using the RAG by considering their applications to a soft-landing problem of a mass–spring–damper system.