Paper Key : IRJ************888
Author: Kapil Rajput ,Saifali Shaikh,Aniket Joshi,Yash Shinde,Guide: Prof. Gauri Kulkarni
Date Published: 04 Apr 2025
Abstract
Reinforcement Learning (RL) has emerged as a powerful paradigm for training agents to make optimal decisions in complex environments. While discrete action spaces have been extensively explored, continuous action spaces present unique challenges due to their infinite dimensionality and the need for efficient exploration strategies. This paper delves into the state-of-the-art techniques for RL with continuous actions, focusing on policy gradient methods, actor-critic architectures, and function approximators. We discuss the advantages and limitations of various approaches, including deterministic policy gradient, proximal policy optimization, and deep deterministic policy gradient. Furthermore, we explore recent advancements in exploration techniques, such as curiosity-driven exploration and intrinsic motivation, to address the challenge of efficiently sampling from high-dimensional continuous action spaces. Finally, we highlight potential research directions and open challenges in the field, such as handling sparse rewards, transferring knowledge between tasks, and ensuring safety and robustness in real-world applications.
DOI Requested