Dynamic order picking has usually demonstrated significant impacts on production efficiency in warehouse management. In the context of an automotive-part warehouse, this paper addresses a dynamic multi-tour order-picking problem based on a novel attention-aware deep reinforcement learning-based (ADRL) method. The multi-tour represents that one order-picking task must be split into multiple tours due to the cart capacity and the operator’s workload constraints. First, the multi-tour order-picking problem is formulated as a mathematical model, and then reformulated as a Markov decision process. Second, a novel DRL-based method is proposed to solve it effectively. Compared to the existing DRL-based methods, this approach employs multi-head attention to perceive warehouse situations. Additionally, three improvements are proposed to further strengthen the solution quality and generalization, including (1) the extra location representation to align the batch length during training, (2) the dynamic decoding to integrate real-time information of the warehouse environment during inference, and (3) the proximal policy optimization with entropy bonus to facilitate action exploration during training. Finally, comparison experiments based on thousands of order-picking instances from the Swedish warehouse validated that the proposed ADRL could outperform the other twelve DRL-based methods at most by 40.6%, considering the optimization objective. Furthermore, the performance gap between ADRL and seven evolutionary algorithms is controlled within 3%, while ADRL can be hundreds or thousands of times faster than these EAs regarding the solving speed.

This study applies an intersectional approach to address concerns aboutdiversity of data acquisition when applying computer vision systems in humanorder picking. The study draws empirical data from a single case study conductedat an automotive manufacturer. It identifies critical factors of intersectionality forthe use of vision systems to enrich data collection in human order picking at fourlevels including form and function, experience and services, systems and infrastructure,and paradigm and purpose. These findings are helpful for mitigating biasand ensuring accurate representation of the target population in training datasets.The results of our study are indispensable for enhancing human-centricity whenapplying computer vision systems, and facilitating the acquisition of unstructureddata in human order picking. The study contributes to enhancing diversity in humanorder picking, a situation that is highly relevant because of the variations in age,gender, cultural background, and language of staff. The study discusses theoreticalandmanagerial implications of findings, alongside suggestions for future research.