A Comprehensive Survey of Advances in Deep Reinforcement Learning for Autonomous Robotics Applications

Abstract

This survey provides a comprehensive overview of the recent advancements in Deep Reinforcement Learning (DRL) for autonomous robotics applications. It begins by introducing the fundamental concepts of DRL and its significance in enabling intelligent and adaptive robotic behavior. The core of this paper is a structured review of state-of-the-art DRL algorithms, categorized into value-based, policy-based, and actor-critic methods, and their innovative applications across a spectrum of robotic domains. We delve into key areas such as manipulation, locomotion, navigation, and human-robot interaction, highlighting the transformative impact of DRL in solving complex, high-dimensional control problems. Furthermore, the survey critically examines the persistent challenges in the field, including the simulation-to-reality gap, sample efficiency, safety, and reward engineering. Finally, we discuss emerging trends and promising future research directions that are poised to shape the next generation of autonomous robotic systems.