Scholarlink

Markov Decision Processes formalize many problems of interest and have been tackled using a variety of techniques, including Reinforcement Learning (RL) and Model Predictive Control (MPC). While each approach has both advantages and disadvantages, RL and MPC have been very successful in the respective domains. RL makes it possible to obtain optimality for the real system, without the need for a model. MPC requires a model, but makes it possible to provide strict stability and safety guarantees, as well as to promote explainability. In this regard, the two techniques are complementary, and this chapter focuses on how they can be combined in order to leverage the advantages of both.

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

Reinforcement Learning with Guarantees

  • 章节

  • 发表日期：

    2025年

查看全文
  • Mario Zanon,
  • Sébastien Gros

摘要

Markov Decision Processes formalize many problems of interest and have been tackled using a variety of techniques, including Reinforcement Learning (RL) and Model Predictive Control (MPC). While each approach has both advantages and disadvantages, RL and MPC have been very successful in the respective domains. RL makes it possible to obtain optimality for the real system, without the need for a model. MPC requires a model, but makes it possible to provide strict stability and safety guarantees, as well as to promote explainability. In this regard, the two techniques are complementary, and this chapter focuses on how they can be combined in order to leverage the advantages of both.