This research addresses critical limitations in existing AGV path planning algorithms, including computational inefficiency, poor dynamic adaptability, and prolonged response times that restrict real-world industrial transportation optimization applications. The study aims to develop an efficient hybrid algorithm to enhance AGV system performance in complex industrial environments.
The study introduces AG-Dueling DQN (A* Guided Dueling Deep Q Network), a novel hybrid algorithm that combines A* heuristic search with Dueling DQN's value function decomposition architecture. The methodology leverages A* to rapidly identify optimal path regions, substantially reducing Dueling DQN's exploration state space while maintaining learning stability through separate state-value and action-advantage estimation. Experimental validation was conducted across diverse workshop configurations.
Experimental results demonstrate that AG-Dueling DQN achieves optimal path identification with significantly reduced training time compared to conventional Dueling DQN. The algorithm exhibits superior environmental perception and adaptive response to dynamic conditions. Performance evaluation shows 34% and 68.74% response time reduction relative to A* in small and large-scale environments, respectively, with robust performance across different workshop configurations.
This research presents the first integration of A* heuristic search with Dueling DQN architecture, innovatively addressing exploration efficiency challenges in deep reinforcement learning for AGV path planning. The hybrid approach maintains reinforcement learning adaptability while significantly improving computational efficiency, providing superior practical applicability for complex industrial AGV operations and advancing intelligent logistics automation systems.
