Recent Advances in Fuzzy Reinforcement Learning for Robot Path Planning

Abstract

Path planning is a fundamental technology for enabling autonomous navigation and intelligent operation of robotic systems. Its primary objective is to generate an optimal or near-optimal trajectory while satisfying obstacle avoidance, kinematic, and task constraints. Traditional path planning methods generally rely on accurate environmental models and exhibit limited adaptability in dynamic and uncertain environments. Reinforcement learning (RL) offers a promising paradigm for robot path planning by enabling agents to learn optimal policies through continuous interaction with the environment. However, challenges such as low learning efficiency and slow convergence remain in continuous state–action spaces. Fuzzy reinforcement learning (FRL), which integrates the uncertainty-handling capability of fuzzy logic with the autonomous decision-making ability of reinforcement learning, has emerged as an important research direction in robot path planning in recent years.This paper presents a comprehensive review of the fundamental principles, key techniques, and applications of FRL in robot path planning. First, the theoretical foundations and integration framework of FRL are introduced. Existing studies are then classified into four representative technical categories according to the role of fuzzy logic within the reinforcement learning framework: fuzzy reward shaping, fuzzy policy representation, fuzzy value function approximation, and fuzzy deep reinforcement learning. The characteristics, advantages, and limitations of each category are comparatively analyzed. Subsequently, recent advances in the application of FRL to typical robotic platforms, including mobile robots, unmanned aerial vehicles (UAVs), unmanned surface vehicles (USVs), and wall-climbing robots, are summarized. Finally, current challenges, including training efficiency, model interpretability, and generalization capability, are discussed, and future research directions, such as adaptive fuzzy reinforcement learning, explainable artificial intelligence, digital twin-driven training, and the integration of planning and control, are highlighted.