A survey on autonomous navigation for mobile robots: From traditional techniques to deep learning and large language models

Abstract

Autonomous navigation is a cornerstone of modern robotic systems. This review provides a comprehensive analysis of the landscape of obstacle avoidance and path planning techniques for mobile robots. We categorize and evaluate a range of approaches, beginning with traditional graph-based methods such as A* and Dijkstra, and geometric techniques like Voronoi diagrams and cell decomposition. The review extends to modern metaheuristic algorithms, including genetic algorithms (GA), particle swarm optimization (PSO), and ant colony optimization (ACO). Furthermore, we explore hybrid models that integrate traditional methods with machine learning, such as reinforcement learning (RL) and neural networks (NN). These hybrid approaches aim to address specific challenges, including escaping local minima and enabling real-time decision-making in uncertain environments. A significant focus is placed on the emerging role of Large Language Models (LLMs), analyzing their application in translating natural language commands into navigational actions and improving human-robot interaction. This work critically analyzes the trade-offs of each paradigm—including computational efficiency, scalability, and adaptability across these diverse methods. Finally, this review outlines emerging trends and open challenges, highlighting potential research directions in collaborative robotics, multi-agent systems, and the broader field of mobile robot navigation.