This study introduces deadlock avoidance algorithms for Automated Guided Vehicles (AGVs) within warehouse fulfillment systems. Numerous AGVs are tasked with transporting requests to deliver shelves to workers stationed at various locations. The pathfinding challenge within such systems is identified as the Multi-Agent Pickup and Delivery (problem) and has been extensively explored in computer science and AI. Nevertheless, most prior research was conducted in environments that are discretized and allow circular movements of AGVs, potentially leading to deadlocks in real-world applications. Our proposed algorithm leverages a dynamic path block method and path reservation. Specifically, it limits movement to certain edges when AGVs navigate along reserved paths, while the remaining AGVs plan their routes using unrestricted edges. We show that our algorithm not only effectively prevents deadlocks but also scales well in environments with a high number of AGVs. •The study introduces a deadlock avoidance algorithm for AGVs.•Deadlock-free path planning framework for bidirectional AGV systems is proposed.•The algorithms’ superior performance is demonstrated with simulation results.