Citation: Joo, T.; Jun, H.; Shin, D.
Task Allocation in Human–Machine
Manufacturing Systems Using Deep
Reinforcement Learning.
Sustainability 2022, 14, 2245. https://
doi.org/10.3390/su14042245
Academic Editor: Paolo Renna
Received: 11 January 2022
Accepted: 9 February 2022
Published: 16 February 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Article
Task Allocation in Human–Machine Manufacturing Systems
Using Deep Reinforcement Learning
Taejong Joo
1
, Hyunyoung Jun
2
and Dongmin Shin
2,
*
1
Industrial Engineering & Management Sciences, Northwestern University, Evanston, IL 60208, USA;
tjoo@u.northwestern.edu
2
Industrial and Management Engineering Department, Hanyang University, Ansan 15588, Korea;
hy1911@hanyang.ac.kr
* Correspondence: dmshin@hanyang.ac.kr
Abstract:
Catering for human operators is a critical aspect in the sustainability of a manufacturing
sector. This paper presents a task allocation problem in human–machine manufacturing systems.
A key aspect of this problem is to carefully consider the characteristics of human operators having
different task preferences and capabilities. However, the characteristics of human operators are
usually implicit, which makes the mathematical formulation of the problem difficult. In addition,
variability in manufacturing systems such as job completion and machine breakdowns are prevalent.
To address these challenges, this paper proposes a deep reinforcement learning-based approach
to accommodate the unobservable characteristics of human operators and the stochastic nature of
manufacturing systems. Historical data accumulated in the process of job assignment are exploited to
allocate tasks to either humans or machines. We demonstrate that the proposed model accommodates
task competence and fatigue levels of individual human operators into job assignments, thereby
improving scheduling-related performance measures compared to classical dispatching rules.
Keywords:
deep learning; dynamic task allocation; human factors; intelligent manufacturing systems;
manufacturing scheduling; reinforcement learning
1. Introduction
Although automation has been adopted for repetitive physical tasks in manufacturing
systems, human operators are indispensable components, even in advanced manufacturing
systems. For instance, in automotive door assembly tasks, an operator holds and guides
the door using an assistive material handling apparatus or equipment to attach it to the
main body of the vehicle. These types of hybrid systems can take advantages of both
humans and machines. For physical manufacturing activities, humans offer the inimitable
senso-motoric ability for complex tasks, and machines can achieve higher speed, constant
precision, and force [1,2].
This paper presents a task allocation problem in human–machine manufacturing
systems. It aims to find desirable job allocation to manufacturing resources in such a way
that tasks are assigned to human operators with consideration of their competence and
preference while guaranteeing reasonably good operational performance measures such as
makespan, flowtime, and/or tardiness.
Over the decades, optimization-based techniques have long been a mainstream for
addressing scheduling problems. In doing so, mathematical objective functions and con-
straints are first defined, and then the optimization algorithm is employed to find solutions
of the formulated problem. This optimization-based approach has assumed availability of
explicit formulations concerning objective functions and constraints of the target systems.
Allocating manufacturing tasks to human operators requires careful considerations for
their characteristics and properties. While machines usually perform specific tasks such as
machining process or assembly, the human operators can perform various tasks depending
Sustainability 2022, 14, 2245. https://doi.org/10.3390/su14042245 https://www.mdpi.com/journal/sustainability
