Transaction assignment policy in double-deep SBS/RSs by using Deep Q-learning

The double-deep shuttle-based storage and retrieval system (SBS/RS) has gained more attention due to its higher area utilization. However, the additional relocation operations for blocking loads increase the complexity of transaction assignment (TA) challenges. This paper investigates a real-Time TA method for a double-deep SBS/RS using the Deep Q-Network (DQN), where shuttles have the flexibility to travel between tiers freely. Using the commonly applied Nearest Free relocation strategy, the system acts as an agent to learn the optimal policy in real-Time. It selects a retrieval task from the transaction pool, pairs it with a storage task to create a dual command cycle, and assigns it to an available idle shuttle for execution in the next cycle. The effectiveness of the proposed DQN method was validated through experimental comparisons with various static TA strategies.