Model-Based Reinforcement Learning: Challenges, Methods, and Progress

EasyChair Preprint 11785

Abstract

This paper presents a comprehensive survey of model-based reinforcement learning (MBRL), a prominent paradigm in artificial intelligence and machine learning. Model-based reinforcement learning aims to enhance the efficiency and sample complexity of learning by leveraging explicit models of the environment. The survey delves into the challenges faced by MBRL approaches, ranging from model inaccuracies to computational complexity. It provides a thorough examination of various methods employed in MBRL, including model learning, planning, and policy optimization techniques. The paper also highlights the significant progress made in recent years, showcasing innovative advancements and successful applications in diverse domains. By offering insights into the state-of-the-art methodologies, the survey contributes to a deeper understanding of the current landscape, paving the way for future developments in model-based reinforcement learning.

Keyphrases: Reinforcement Learning, based, model

@booklet{EasyChair:11785,
  author    = {Kurez Oroy and Robert Thomson},
  title     = {Model-Based Reinforcement Learning: Challenges, Methods, and Progress},
  howpublished = {EasyChair Preprint 11785},
  year      = {EasyChair, 2024}}