Robotic Systems

Chair: Ivo Adan (Eindhoven University of Technology)

Title: Human and Robot Collaboration in Warehouses
Speaker: Alexandros Pasparakis (Erasmus University Rotterdam)

Authors: Alexandros Pasparakis, Jelle De Vries, René De Koster, Rotterdam School of Management, Rotterdam, Netherlands. Contact:

Abstract: Robotic material handling technologies are being increasingly adopted in warehouses. Automated guided vehicle (AGV) systems supporting the order picking process, called pick support AGVs, are one such example. By combining the human flexibility in performing complex tasks with the robot tirelessness in performing repetitive tasks, human and robot collaboration (HRC) is disrupting the ways manual order picking is traditionally performed. In order to prepare for this new reality, we conducted an experiment to compare multiple scenarios of HRC settings (picker leading vs AGV leading), and investigate both objective and subjective outcomes of HRC: (i) how do order pickers perform (order picking productivity and accuracy) in different HRC settings, as well as (ii) which behavioral implications this robotized environment implies for order pickers (user adoption behavior, task satisfaction, self-evaluation). Results indicate that when pickers are leading AGVs (compared to when pickers are following AGVs), they perform better in terms of order picking productivity, but worse in terms of order picking accuracy. Furthermore, we find that high levels of picker prevention regulatory focus may bridge the productivity gap between picker-leading vs robot-leading settings. Evidence that that HRC positively affects picker task satisfaction and self-evaluation is also established.

Title: Large fork-join networks with nearly deterministic service times
Speaker:  Dennis Schol (Eindhoven University of Technology)

Supervisors: Bert Zwart & Maria Vlasiou

Abstract: A typical property of high-tech manufacturing is that a large number of suppliers are involved, and are specialized in producing and delivering a very specific component of the final product. This study is motivated by modeling the delays in the emerging complex supply chain. Consider such a chain with many suppliers, each producing a specific component of the final product. In this system, the delay of the manufacturer is determined by the slowest supplier.

To model this delay, we propose a discrete time N server fork-join queueing network, in which each server represents a unique supplier. Each time step we have one arrival with a given probability p, and each server completes one service with probability q, independent of the other servers.  These p and q are close to 1, thus we have nearly deterministic arrivals and services. The aim of this study is to approximate the length of the largest of the N queues in the network.

We present a fluid limit and a steady-state result for the maximum queue length, as N goes to infinity. In order to get the fluid limit, we have to scale time with N3 and space with 1/N. For the steady-state result, we only have the spatial scaling of 1/N. These two results have remarkable differences. The steady-state result depends on two model parameters, while the fluid limit only depends on one model parameter. Moreover, we see that the maximum queue length has a different scaling order in steady state compared to the fluid limit.

In order to prove these results, extreme value theory and diffusion approximations for the queue lengths are used. Since each queue has the same arrival process, queue lengths are dependent random variables, which makes it challenging to find convergence results of the maximum queue length. By giving upper and lower bounds on the maximum queue length, each having a small dependent and a large independent part, we are able to prove these convergence results. Finally, we do some numerical analysis to determine the phase transition between the fluid limit and the steady state result.

Title: Correlated Dispersed Storage Assignment in Robotic Warehouses
Speaker: Masoud Mirzaei (Eindhoven University of Technology)

Abstract: Warehouses, particularly in ecommerce retail, use automation to meet the customer expectations of short order throughput times. Order picking time is a key component of order throughput time. The storage assignment policy, which defines where the products are stored in the warehouse, is a main determinant of order picking time. Random storage is a simple and straightforward policy. A product turnover frequency-based storage policy, where products with higher turnover frequency are stored closer to the depot, typically leads to shorter throughput times. However, these common storage policies neglect the information on product correlation in customer demand. For larger orders picked from large assortments (or when small orders are picked in batch), it may pay off to also consider correlated storage assignments. Items appearing jointly in such orders can be stored on the same storage pod such that fewer pods should be retrieved to pick the order. Additionally, the inventory of a product can be split and assigned to multiple storage pods to increase the correlation of products on different pods. We formulate a mathematical model for the correlated dispersed assignment (CDA) that clusters the correlated products on the pods and assigns the clusters to storage zones to minimize the total retrieval time. An efficient heuristic is developed to solve large instances of the problem. A data set of a cosmetic products retailer is used for analysis. The numerical experiment shows that the CDA offers a considerable improvement compared to other storage policies. Statistical models show that the correlation has a significant role in the performance of the CDA.

In cooperation with:

Nima Zaerpour (College of Business Administration, California State University San Marcos, the United States)
René de Koster (Rotterdam School of Management, Erasmus University, Rotterdam, the Netherlands)