Analysing the Effects of Congestion on Hybrid Order Picking Systems Using a Discrete-Event Simulator

Charlie Street, Sri Sadhan Jujjavarapu, Michael Chen, Sanjoy Paul, Nick HawesLecture Notes in Networks and Systems 2024 · DOI · OpenAlex

Abstract

This study investigates the impact of human-robot congestion on the performance and economic viability of hybrid Order Picking Systems (OPS). Utilizing a Discrete-Event Simulator (CAMAS), the research quantifies how traffic and queuing congestion lead to diminishing returns as the fleet size of Autonomous Guided Vehicles (AGVs) increases. Results indicate that while automation can double throughput in ideal conditions, heavy congestion can render additional robots counterproductive, highlighting the need for strategic fleet management and layout considerations in modern logistics.

Introduction

In modern logistics, Order Picking Systems (OPS) serve as the critical backbone of warehouse operations. Traditionally, these systems relied entirely on manual labor. However, the industry is rapidly transitioning toward hybrid OPS, where human pickers and Autonomous Guided Vehicles (AGVs) share the same workspace to fulfill orders.

Hybrid warehouse environment featuring (a) human pickers and (b) autonomous guided vehicles (AGVs) sharing the same workspace.

A primary challenge in these environments is congestion. When multiple agents—human or robotic—occupy narrow aisles, they must slow down to avoid collisions (traffic congestion) or wait for access to specific picking locations (queuing congestion). While traffic flow is well-studied in highway engineering, the complex, interactive "dance" between humans and robots in a warehouse is significantly harder to model. This paper aims to quantify the performance degradation caused by congestion and determine the optimal fleet size for hybrid systems.

Methodology

The researchers employed a Discrete-Event Simulator (DES) called CAMAS. Unlike high-fidelity physics simulators that are computationally expensive, a DES allows for the rapid testing of thousands of scenarios by focusing on high-level events, such as a robot arriving at a specific shelf.

Congestion Modeling

To capture the nuances of warehouse traffic, the study utilized context-aware duration distributions. Instead of assuming a fixed travel speed, travel times were modeled as probability distributions that shift based on the number of agents present on a path. The researchers tested five distinct models (Model 0 to Model 4), ranging from zero traffic interference to heavy congestion penalties.

Task Allocation

Orders were assigned to pickers and AGVs using a Sequential Single-Item (SSI) auctioning mechanism. To prevent operational deadlocks, the system prioritized orders based on the human picker's schedule, ensuring that robots did not block human workflows unnecessarily.

Experimental Setup

The study utilized a simulated environment based on a realistic warehouse layout. The topological map included 6 zones, 12 racks, and 8 bays per rack, with 3 dedicated packing locations.

Topological map of the simulated warehouse layout, divided into 6 zones with 12 racks and 3 packing locations.

To ensure realism, the DES was calibrated using empirical navigation data collected from Emulate3D, a physics-based simulator that employs anthropometric models to simulate realistic human movement. Evaluations were performed over 10 repeats of a mission containing 600 randomly generated orders.

Key Insights and Results

The Throughput Plateau

The simulation revealed a clear "throughput plateau." In an ideal scenario with no congestion, adding more AGVs consistently improves performance. However, in realistic models, the time saved by automation is eventually offset by the time agents spend waiting in crowded aisles. Under heavy congestion (Model 3), a 10-AGV fleet provided only a 3% increase in throughput, compared to a 105% increase in a zero-congestion scenario.

The Net Negative Threshold

In the most extreme congestion model (Model 4), adding any number of AGVs actually decreased throughput compared to a purely manual system. This suggests that in cramped environments, automation without corresponding layout changes can be a detrimental investment.

Economic Impact and Optimal Fleet Size

The "optimal" number of robots depends heavily on the investment timeline. Given an upfront cost of approximately $30,000 per AGV, a 1-year profit analysis favors a smaller fleet (around 10 robots). However, a 5-year analysis indicates that 20–30 robots provide the best return on investment for medium congestion levels. Beyond 30 robots, queuing at packing stations becomes the primary bottleneck, and the costs of maintenance and congestion begin to erode total profit.

Conclusion and Future Work

This research demonstrates that congestion is a non-negligible factor in warehouse automation. While AGVs offer significant potential for throughput gains, these gains are subject to diminishing returns. Future work will address current limitations, such as the assumption that AGVs carry only one order at a time and the exclusion of human factors like fatigue. Additionally, investigating how dynamic warehouse layouts (e.g., adjustable aisle widths) might mitigate congestion remains a high-priority area for further study.

Summary updated

BibTeX
@inproceedings{street2024analysing,
  author = {Charlie Street and Sri Sadhan Jujjavarapu and Michael Chen and Sanjoy Paul and Nick Hawes},
  title = {{Analysing the Effects of Congestion on Hybrid Order Picking Systems Using a Discrete-Event Simulator}},
  booktitle = {Lecture Notes in Networks and Systems},
  year = {2024},
  doi = {10.1007/978-3-031-44851-5\_30},
}
Figure