Review on Optimizing Robotic Navigation with Deep Reinforcement Learning Algorithms

Abstract

Robotic path planning is a critical aspect of autonomous robot navigation, enabling robots to efficiently navigate in complex environments while avoiding obstacles and reaching their intended destinations. Traditional path planning algorithms often struggle with intricate and dynamic environments due to their reliance on predefined maps assumptions about the environment's behavior. In recent few years, deep reinforcement learning (DRL) has come up as a promising approach for enhancing robotic path planning. RL techniques allow robots to learn optimal or near-optimal paths through trial-and-error interactions with their surroundings, adapting to changing environments and unforeseen obstacles. This review paper provides overview of the progress in enhancing robotic path planning using reinforcement learning. We categorize the existing research depending on the types of RL algorithms employed, such as Q-learning, policy gradients, and actor-critic methods, among others. By synthesizing recent research findings, this review paper offers insights into the current state of enhancing robotic path planning using reinforcement learning, identifies open challenges, and suggests potential directions for future research. The synergy between RL and robotic path planning holds great promise for revolutionizing autonomous navigation, enabling robots to travel in complex and dynamic environments with unprecedented efficiency and adaptability.