Reinforcement Learning Techniques for Autonomous Robotics

Abstract

This paper examines reinforcement learning (RL) methods for autonomous robots and their strengths, weaknesses, and applications. The main goals are to assess sophisticated RL algorithms in robotics, identify problems, and suggest improvements. This secondary data-based review synthesizes current research on Deep Q-networks (DQN), policy gradient techniques, model-based approaches, and hierarchical RL. These strategies improve robotic learning by boosting sample efficiency, managing continuous actions, and enhancing real-time performance. Still, they also confront sim-to-real gaps, safety issues, and high computing demands. The paper recommends investing in simulation-to-reality transfer research, safety measures, and computational tools to solve these constraints. The study emphasizes the revolutionary potential of RL in autonomous robots and the need for continuing innovation and supporting policy to overcome limitations and fully harness RL capabilities in practical applications.