European Journal of Operational Research 277 (2019) 396–411

Contents lists available at ScienceDirect

European Journal of Operational Research

journal homepage: www.elsevier.com/locate/ejor

Invited Review

Warehousing in the e-commerce era: A survey

Nils Boysen a,1,∗, René de Koster b,2, Felix Weidinger a,1
a
Lehrstuhl für Operations Management, Friedrich-Schiller-Universität Jena, Carl-Zeiss-Str. 3, Jena 07743, Germany
b
Rotterdam School of Management, Erasmus University Rotterdam, P.O. Box 1738, DR Rotterdam 3000, The Netherlands

a r t i c l e i n f o a b s t r a c t

Article history: E-commerce retailers face the challenge to assemble large numbers of time-critical picking orders each
Received 11 April 2018 consisting of just a few order lines with low order quantities. Traditional picker-to-parts warehouses are
Accepted 14 August 2018
often ill-suited for these prerequisites, so that automated warehousing systems (e.g., automated picking
Available online 23 August 2018
workstations, robots, and AGV-assisted order picking systems) are applied and organizational adaptions
Keywords: (e.g., mixed-shelves storage, dynamic order processing, and batching, zoning and sorting systems) are
Logistics made in this branch of industry. This paper is dedicated to these warehousing systems especially suited
Warehousing for e-commerce retailers. We discuss suited systems, survey the relevant literature, and define future
E-commerce research needs.
Survey
© 2018 Elsevier B.V. All rights reserved.

1. Introduction • Small orders: Most private consumers order rather few order
lines each demanding only very few items. At German Amazon
Warehousing, i.e., the intermediate storage of goods in between warehouses, for instance, the average order demand amounts to
two successive stages of a supply chain (Bartholdi & Hackman, merely 1.6 items (Boysen, Stephan, & Weidinger, 2018c).
2016), and its basic functions, i.e., receiving, storage, order pick- • Large assortment: Items offered on websites consume no costly
ing, and shipping (Gu, Goetschalckx, & Ginnis, 2007), are essen- storage space in stores and are, nonetheless, accessible to a
tial components in any supply chain. Therefore, it is not surpris- broad public. That is why online retailers can afford a much
ing that a vast body of literature on this topic has accumulated larger assortment, and niche products, typically, account for a
over the past decades (see the most recent surveys of Gu et al. much larger proportion of sales in e-commerce than they do in
(2007), Gu, Goetschalckx, and McGinnis (2010), de Koster, Le-Duc, brick-and-mortar stores. This phenomenon is also known under
and Roodbergen (2007), and Azadeh, de Koster, and Roy (2018)). the term “the long tail” (Brynjolfsson, Hu, & Smith, 2003).
According to de Koster et al. (2007), in 2007, more than 80% of all • Tight delivery schedules: Next-day or even same-day deliveries
warehouses in Western Europe still followed the traditional picker- are an elementary promise of many online retailers especially
to-parts setup. Here, human pickers pick order after order while in the B2C segment (e.g. Yaman, Karasan, & Kara, 2012). This
successively visiting shelves on which the demanded stock keeping puts increasing stress on warehouse operations and leads to
units (SKUs) are stored. The major drawback of these traditional highly time critical order fulfillment processes.
warehouses is the unproductive picker walking when moving from • Varying workloads: Many online retailers have dynamically ex-
shelf to shelf and back to the central depot. panded their warehouse capacities over the past years (Laudon
The ever increasing sales volumes of e-commerce in the past & Traver, 2007) and face highly volatile demands, depending on
decade (e.g., see Statista, 2017), however, gave rise to a new gener- the offered products, e.g., due to seasonal sales. Thus, scalable
ation of warehouses, which are specifically tailored to the special warehouse capacities are required that can flexibly be adjusted
needs of online retailers directly serving final customer demands in to varying workloads.
the business-to-consumer (B2C) segment. They, typically, face the
following requirements:
Conventional warehouses have difficulties meeting these re-
quirements. In a traditional pick-by-order, picker-to-parts system
∗
a picker starts and ends each tour collecting a pick list in a cen-
Corresponding author.
tral depot. If orders are small, the fraction of unproductive work
E-mail addresses: nils.boysen@uni-jena.de (N. Boysen), rkoster@rsml.nl (R. de
Koster), felix.weidinger@uni-jena.de (F. Weidinger). while walking to and from the depot and between shelves is over-
1
http://www.om.uni-jena.de. proportionally large. If not counterbalanced by a large (and costly)
2
https://www.rsm.nl. workforce or by batching multiple orders, the resulting loss of

https://doi.org/10.1016/j.ejor.2018.08.023
0377-2217/© 2018 Elsevier B.V. All rights reserved.
 N. Boysen et al. / European Journal of Operational Research 277 (2019) 396–411 397

Fig. 1. Overview of warehousing systems suited for e-commerce.

throughput makes it hard meeting the tight delivery schedules of We, thus, define a warehousing system as a hardware or process
online retailing. element (or a combination of both) that enables the intermedi-
To avoid these problems of traditional warehouses, novel ware- ate storage of goods in between two successive stages of a supply
housing systems have been developed that either apply automa- chain (Bartholdi & Hackman, 2016).
tion or implement organizational adaptions. They have a much Online retailing, typically, has to assemble (i) small orders from
better fit for online retailers in the B2C segment. This paper is (ii) a large assortment under (iii) great time pressure and has to
dedicated to surveying these warehousing systems from an oper- flexibly adjust order fulfillment processes to (iv) varying workloads
ational research perspective. We introduce these systems, discuss (see Section 1). Clearly, the importance of these four requirements
their suitability for e-commerce, describe the basic decision prob- varies, e.g., with the offered products and the business strategy,
lems to be solved when setting up and operating each system, and and there are special cases where other requirements may be even
review the relevant literature. Furthermore, we define future re- more important, e.g., the flexibility to also handle large orders of
search needs in this area. brick-and-mortar stores in an omni-channel sales strategy (Hübner,
For this purpose, the remainder of the paper is structured as Holzapfel, & Kuhn, 2016). For special products some of these re-
follows. Section 2 defines the scope of our survey by identify- quirements may even not hold at all. In online grocery retailing,
ing the investigated warehousing systems relevant for e-commerce for instance, orders containing dozens of order lines are rather
and by specifying the structure of our survey. Then, Sections 3–8 the norm (Valle, Beasley, & da Cunha, 2017). We, however, exclude
are each dedicated to a specific warehousing system and, finally, these special cases and presuppose that requirements (i) to (iv) are
Section 9 concludes the paper by discussing some general research of outstanding importance in online retailing. It is part of our sur-
directions in warehousing. vey to discuss the suitability of each introduced warehousing sys-
tem for these requirements.
2. Scope and structure This leads us directly to the question, which specific warehous-
ing systems fit these requirements and should, thus, be consid-
This survey is dedicated to warehousing systems that are well ered in this survey. Unfortunately, deriving objective selection cri-
suited for the special requirements of online retailers in the B2C teria seems barely possible, so that we decide for the following
segment. A warehousing system consists of hardware, which can be approach. First, we preselected the list of warehousing systems de-
further subdivided into storage devices (e.g., a rack), material han- picted in Fig. 1 (ordered according to their level of automation),
dling systems (e.g., a conveyor belt) and picking tools (e.g., a pick- which is based on numerous warehouses visits, interviews with
by-voice solution or a picking workstation), and processes defining warehouse managers and experts, and the websites of manufac-
the work flow along the applied hardware elements. The critical el- turers of warehousing solutions. Then, the authors rated whether
ement which makes a warehouse system suited for online retailing or not the respective warehousing system fulfills requirements (i)
can be either related to hardware or processes. Typically, however, to (iv) on a binary scale and discussed the plausibility of the out-
a combination of multiple hardware innovations and process ele- come with warehousing experts (scientists and consultants). The
ments is applied, so that we do not further differentiate hardware result is depicted in Fig. 1 where a system fitting the respective
and processes, but refer to most important entities in both areas. requirement is indicated by the check icon. Based on these results
398 N. Boysen et al. / European Journal of Operational Research 277 (2019) 396–411

we review all warehousing systems fulfilling at least three out of (also denoted as put-to-light systems), see (Füßler & Boysen,
four requirements. Consequently, we exclude the following ware- 2017a). These systems are not treated in this survey.
housing systems from our survey: • We also exclude systems dedicated to specific product types
that require specific handling, e.g., due to their weight or di-
• For traditional reasons, quite a few online retailers still mension. Examples are, for instance, hanging trolley solutions
use their conventional picker-to-parts warehouses from pre- for foodstuffs (Swisslog, 2016) and clothes or crane-based sys-
internet times where unit loads are stored in racks and order tems for tires (Cimcorp, 2011).
after order is picked and brought to a central depot. Due to the • Finally, prototype systems that have not outgrown the eval-
large fraction of unproductive walking (or driving) time, how- uation phase and are yet not applied by large online re-
ever, these traditional systems can reach the tight deadlines of tailers are also excluded. Examples are the Toru robot of
online retailing only at the cost of an excessively large (and ex- Magazino (2017) or the Robo-Pick system of Schäfer (2017),
pensive) workforce. Although these warehouses are still applied which promise fully automated order picking from shelves and
for online retailing, we rather focus on systems specifically de- bins, respectively.
signed for the aforementioned requirements. Note that exten- Excluding these warehousing systems leaves behind the six sys-
sions of this basic setup, e.g., batching of orders, have a better tems that are discussed in the following sections ordered by an
fit and are, thus, considered in our survey. increasing level of automation, e.g., in the same order depicted in
• In the recent years, quite a few automated compact storage Fig. 1. For each system we elaborate the following aspects:
systems have been developed. Examples are movable rack sys-
tems (Boysen, Briskorn, & Emde, 2018b), shuttle-based deep (a) System description: We give a brief introduction of the main
lane storage systems (Boysen, Boywitz, & Weidinger, 2018a; Za- system elements and their involvement in the order fulfill-
erpour, Yu, & de Koster, 2015), the live-cube system (Zaerpour, ment process.
Yu, & de Koster, 2017), and puzzle-based storage systems (Gue (b) Suitability for e-commerce: Given our four basic require-
& Kim, 2007). A detailed overview on compact storage is pro- ments of e-commerce warehouses (see Section 1), we briefly
vided by the survey paper of Azadeh et al. (2018). These sys- summarize the fit of each system.
tems use space efficiently by storing products closer together (c) Decision problems: For each system, basically the following
without providing a direct access to each single item (at any three main decision areas exist (ordered according to plan-
time). With automation and intelligent planning approaches, ning horizon): (i) layout design planning, (ii) storage assign-
these systems aim to, nonetheless, ensure acceptably fast re- ment, and (iii) order picking. We discuss what peculiarities
trieval times. In spite of these efforts, compact storage is, typ- have to be considered when addressing these planning tasks
ically, not suited for the tight delivery schedules and vast as- for the respective warehousing system.
sortments of today’s online retailing. Seeing the recent trends, (d) Literature survey and future research: Finally, we also serve
however, where online retailers offer premium delivery services the main intention of a survey paper, that is we review the
within the next few hours, the stored products have to move relevant literature and outline future research needs.
closer to the customers where storage space is rare and costly. The fundamental functionality of warehousing makes it any-
Thus, for these express services compact storage directly in the thing but astounding that quite a few survey papers on warehous-
city centers may be an adequate choice in the future. We rather ing already exist. First, there are the general surveys of van den
aim at the status quo and exclude compact storage systems Berg (1999), Rouwenhorst et al. (20 0 0), Gu et al. (2007, 2010), and
from this survey. de Koster et al. (2007). These papers address the complete field of
• Fully automated A-Frame systems process up to 750,0 0 0 units warehousing, but mostly focus on traditional picker-to-parts sys-
per day at an order accuracy of more than 99.95% (Caputo & tems. Since their publication, some time has passed and online re-
Pelagagge, 2006). An A-Frame consists of a set of successive tailing is still a comparatively novel phenomenon that has dynam-
vertical channels filled with stockpiles of SKUs. The channels ically evolved in the last decade, so that there is not much overlap
are arranged along two opposite frames, which have their top with these general surveys. Moreover, other surveys on special
ends tilted toward each other like a pitched roof. In between warehousing aspects, e.g., automated storage and retrieval system
the two frames runs a conveyer belt, so that from the side view (Gagliardi, Renaud, & Ruiz, 2012; Roodbergen & Vis, 2009) and
the system looks like an “A”. At the bottom of each channel crane scheduling in these systems (Boysen & Stephan, 2016) ex-
there is an automated dispenser, which pushes one or more ist. Furthermore, Azadeh et al. (2018) focus on robotized warehous-
bottommost items toward the conveyor whenever the respec- ing systems. From those systems reviewed in our paper only AGV-
tive items are required by an order passing by on the conveyor. assisted order picking (see Section 6) and shelf-moving robots (see
Due to their efficiency A-Frames are certainly well suited to Section 7) fall into this category. Mostly, Azadeh et al. (2018) ad-
fulfill the tight deadlines of online retailing and also small or- dress compact storage systems, so that there is not much over-
der sizes seem unproblematic. However, large assortments re- lap with their paper. The recent survey of van Gils, Ramaekers,
quire an excessive number of A-Frame modules, which leads Caris, and de Koster (2018) is dedicated to holistic warehousing re-
to high investment costs and excessive space requirements on search that combines two or more planning tasks. Holistic models
the shop floor. A-Frames also require a lot of fixed installed are without doubt an important matter especially from the practi-
machinery, so that a flexible adjustment to varying workloads tioner’s perspective (see Section 9), but lead to a completely dif-
seems problematic. In addition, they have strict requirements ferent focus. Gong and de Koster (2011) address methodological
on product shape, size, and packaging. Therefore, A-Frames are aspects and focus stochastic models. Finally, there is the survey
rather suited for niche online retailers with a reduced assort- of Agatz, Fleischmann, and van Nunen (2008), which addresses e-
ment of relatively small and infrangible products, e.g., pharma- fulfillment in a multi-channel environment. The warehousing as-
ceutics and cosmetics (Bartholdi & Hackman, 2016; Pazour & pect is only briefly covered in this survey. Thus, we think that due
Meller, 2011) and, thus, excluded from our survey. to the great relevance of online retailing and the deviating topics
• Some systems are rather dedicated to processing larger cus- of all these previous papers, yet another survey on warehousing
tomer orders consisting of multiple order lines and/or multi- seems well justified.
ple requested items per line, e.g., placed by brick-and-mortar Finally, we briefly specify our database search for retrieving the
stores. Examples are, for instance, inverse order picking systems papers to be reviewed (see, e.g., Hochrein and Glock (2012) for
 N. Boysen et al. / European Journal of Operational Research 277 (2019) 396–411 399

without excessive automation, so that by adding or removing pick-

ers an adaption to varying workloads is easily possible. The large
assortments of the B2C segment seem also unproblematic as long
as there is enough space in the warehouse for placing racks.
On the negative side, scattered storage causes additional effort
during the put-away of units. Logistics workers have to visit mul-
tiple storage positions instead of just a single one when putting
a complete unit load into storage. Moreover, orders demanding
larger order quantities of some SKU cause problems. In this case,
a picker has to visit multiple storage positions of the requested
SKU until enough units are collected. However, in the B2C segment
SKUs are seldom ordered in large quantities, so that mixed-shelves
storage seems well suited to this field of application.
During the layout design phase of a mixed-shelves warehouse
especially the placement of shelves, aisles, and depots have to be
Fig. 2. Mixed-shelves in a scattered storage warehouse. considered and the support equipment of pickers has to be se-
lected. To use space efficiently, most warehouses place the man-
high racks in multi-story mezzanine systems. While layout design
a general description of how to set up a systematic literature re- is a widely elaborated field of research for traditional picker-to-
trieval). As keywords specifying the business function we apply parts warehouses (e.g., see Baker & Canessa, 2009; Roodbergen
“warehousing”, “distribution center”, “order fulfilment”, and “or- & Vis, 2006), scientific decision support for mixed-shelves ware-
der processing”. Furthermore, a second group of keywords, i.e., “e- houses, e.g., on how to position mixed-shelves or where to place
commerce”, “online retailer”, “online retailing”, and “e-fulfillment” middle aisles and depots, is yet not existent. In mixed racks spe-
is applied to specify the branch of industry. Any combination of cific units cannot be found without IT support, so that pickers need
keywords from the first and second group has been applied as a to be equipped with a handheld scanner or some other device (e.g.,
query in two scholarly databases, namely Business Source Premier a pick-by-voice solution) directing the way toward the picking po-
and Scopus. Additional queries were executed with the names of sitions. Furthermore, in most mixed-shelves warehouses, pickers
the six reviewed warehousing systems. All English-language pa- are equipped with picking carts, which have a capacity for mul-
pers published in peer-reviewed journals that have been retrieved tiple bins, so that multiple orders can be picked in parallel. Prop-
and those cited in their reference lists (snowball approach) were erly dimensioning these carts according to the basic trade-off be-
checked for relevance by analyzing their abstracts. Additionally, tween additional capacity and reduced maneuverability and vice
some selected working papers and conference proceedings have versa seems an interesting task for future research.
been integrated, if (according to the authors’ subjective assess- Like in any other warehouse, the storage assignment decides on
ment) they considerably contribute to the surveyed field. Note that the exact storage position of each unit to be stored. All mixed-
we try to build up on the work of existing survey papers as far as shelves warehouses known by the authors just apply random stor-
possible to avoid repetition. Batching and zoning, for instance, is age and leave the selection of storage positions to the logistics
surveyed in detail by Gu et al. (2007) and de Koster et al. (2007), workers replenishing the shelves. Equipped with a cart full of units
so that we limit our literature review on both topics to work pub- to be stored as well as a handheld scanner or a comparable device,
lished after these surveys have appeared. The same holds for the these workers walk into the warehouse, place units just where
routing of automated guided vehicles, which has been summarized they find suited storage space, and register the new storage po-
by Qiu, Hsu, Huang, and Wang (20 02) and Vis (20 06), such that we sition of the respective SKU in the IT system making use of their
only discuss literature published after 2006 here. device. In the long run, random storage leads to an equal distri-
bution of units all over the warehouse. Note that one could ar-
3. Mixed-shelves storage gue that the human selection makes the choice not actually ran-
dom, but leads to most convenient racks within easy reach being
Mixed-shelves storage is a special storage assignment strategy filled first. Nonetheless, we follow the existing literature and call
applied by many large-sized facilities in the B2C segment, such as this a random storage assignment policy. Weidinger and Boysen
the distribution centers of Amazon Europe and fashion retailer Za- (2017) show that planning the storage positions such that the scat-
lando (Weidinger, Boysen, & Schneider, 2017). Incoming unit loads teredness of units is optimized, may lead to a much higher picking
of SKUs are purposefully broken down into single units that are performance. They operationalize the scatteredness by minimizing
spread all over the shelves throughout the warehouse. This is why the weighted sum of maximum distances toward the closest unit
this storage assignment strategy is also denoted as scattered stor- of each SKU seen from the access points to the central conveyor
age (Weidinger & Boysen, 2017). An example warehouse is depicted system. They develop a heuristic binary search based solution pro-
in Fig. 2.3 cedure and compare the results with random storage. By simulat-
The basic intention of mixed-shelves storage is that with units ing picker tours through both alternative storage plans they show
scattered all around the warehouse the average distances from that optimization leads to much better results. Their solution pro-
anywhere in the warehouse toward the closest unit of each SKU cedure is able to handle a few hundred SKUs. Seeing the huge fa-
are considerably reduced. This increases the probability that a de- cilities and thousands of SKUs in the B2C segment, however, future
manded SKU is close by. Furthermore, mixed-shelves warehouses research could try to develop even better solution procedures that
often have multiple access points (depots) where completed or- can handle data instances of real-world size.
ders are handed over to the central conveyor system. In this way, The main decision problems to be solved during order picking
the unproductive walking of pickers considerably decreases and are the prioritization of orders, their assignment to pickers and
tight delivery schedules can better be met. Scattered storage comes picker routing. The prioritization mainly depends on the urgency
of orders, which, e.g., is influenced by the value of the customer,
3
The picture is published under the Creative Commons License (BY 2.0). The au- promised delivery dates, and the departure times of the targeted
thor of the picture is Álvaro Ibáñez. delivery trucks. To the best of the authors’ knowledge research on
400 N. Boysen et al. / European Journal of Operational Research 277 (2019) 396–411

this topic in the mixed-shelves context is not yet existent. The way, the pick density per tour is increased and a more efficient
assignment of orders to pickers has to consider fairness aspects picking process is enabled.
(i.e., the workload should be evenly shared among pickers) and the • Zoning: A further reduction of the picking effort is enabled, if
fit of orders assigned to the same picker, such that short picking the warehouse is partitioned into disjoint zones. Order pickers
routes are enabled. The assignment decision is, thus, heavily in- only pick the part of an order that is stored in their assigned
fluenced by the picker routing, which decides on the sequence of zone. Parallel zoning enables a parallel processing of orders
visited storage positions where units are retrieved. Scattered stor- and, thus, a faster order processing. Furthermore, each picker
age warehouses cause some peculiarities, which are not consid- only traverses smaller areas of the warehouse. Note that a pro-
ered by the traditional routing methods tailored to unit-load ware- gressive zoning where orders visit zones subsequently seems
houses (see de Koster et al. (2007)). First, the routing of pickers problematic when facing the tight deadlines of online retailers
also requires a selection of alternative storage positions each re- (see below).
quested SKU is stored at. Once a predecessor order has depleted
Either only one of these policies is individually applied or both
a storage position these units are no longer available for succes-
of them are combined. In either case, a major drawback of these
sor orders. Moreover, many mixed-shelves warehouses have multi-
policies is that they require a subsequent consolidation process
ple depots and apply carts with bins for multiple orders, so that
where picking orders are isolated. Batched orders need to be sep-
further peculiarities need to be considered. Existing research in
arated and a parallel zoning requires the merging of multiple par-
this context limits itself to single picker routing. Daniels, Rum-
tial orders picked in different zones. Note that the latter can be
mel, and Schantz (1998) were the first to integrate the selection of
avoided if zoning (in isolation) is applied in a sequential man-
units from alternative storage positions into picker routing. They
ner (also denoted as progressive zoning, see Yu & de Koster, 2009).
prove NP-hardness for facultative distance matrices and develop
In this case, bins in which orders are collected visit zones subse-
multiple heuristics. Among them a tabu search approach is shown
quently, so that an additional consolidation process is not required.
to have the best solution performance. Weidinger (2018) extends
Seeing the great time pressure of online retailing, however, sub-
their findings to rectangular warehouses with parallel aisles by
sequent visits of multiple zones, typically, require too much time,
proving NP-hardness for this special layout setting, which is rather
so that we restrict our view on parallel zoning (see also Petersen,
standard in business practice. He also provides a new decomposi-
20 0 0). Here, partial orders are simultaneously collected in multi-
tion approach for solving the picker routing problem in rectangular
ple zones and, thus, need to be consolidated afterwards. The need
warehouses by combining multiple selection rules and the efficient
for consolidation separates the complete order fulfillment process
dynamic programming approach of Ratliff and Rosenthal (1983). Fi-
of warehouses applying batching and/or zoning into the following
nally, this study provides insight into the disadvantage of mixed-
three steps:
shelves storage when processing orders demanding many units of
the same SKU and gives managerial insights into an appropriate (a) Order picking is executed by human order pickers that col-
level of unit scatteredness for multi-channel strategies. While all lect batches of partial orders in their respective zones. Many
previously mentioned papers deal with the isolated picker routing online retailers equip their pickers with small picking carts
problem, Weidinger et al. (2017) combine picker routing and sort- carrying multiple bins to collect multiple (partial) batches
while-pick batching. The latter means that the picker directly sorts per picker tour in parallel. Once a picker tour is completed,
orders into bins, such that no subsequent consolidation process is the filled bins are handed over to the central conveyor sys-
required. Given multiple customer orders, a given capacity of the tem at an access point (depot) of the respective zone and
picking cart pushed by the picker, and a multi-depot configuration the next tour starts.
of the mixed-shelves warehouse, they search for the shortest pick- (b) Intermediate storage: Completed bins could directly be con-
ing tour satisfying all orders, while picking multiple orders in par- veyed into the consolidation area. Especially in large ware-
allel. Future research could extend the current findings to routing houses, however, the time span between completion of the
multiple pickers in parallel, integrating a parallel replenishment of first and last bin of a batch arriving from different zones
the shelves (see Wruck, Vis, & Boter, 2013), congestion in narrow may become large. For instance, inventory differences or
aisles, and order batching. The latter should integrate time-critical misplaced units can occur, so that some parts of a batch ar-
orders with individual due dates into picker routing, which seems rive considerably delayed. If the bins of such a batch would
highly relevant regarding the trend to ever tighter delivery sched- directly be released into the consolidation area, then, at least
ules. some positions where orders are collected, i.e., some dead-
end lane of a conveyor-based sorting system or the shelves
4. Batching, zoning, and sorting of a put wall (see below), would be blocked for a prolonged
time span until the delayed bins arrive. To avoid the ex-
Two warehousing policies that also have the basic intention cessive consolidation capacity required by such a direct bin
of reducing the unproductive walking of pickers are batching and release (or the threat that all positions are blocked and a
zoning. These policies have a much longer tradition and are ap- deadlock occurs), bins are typically collected in the central
plied in traditional picker-to-parts warehouses (where unit loads conveyor system and only released into the consolidation
are kept together) for decades. In the recent years, large online re- area once the batch is complete and all bins have arrived
tailers like Amazon Europe and fashion retailer Zalando, however, from their zones. Automated (i.e., conveyor-based) consol-
couple batching and zoning with mixed-shelves storage (Weidinger idation systems often apply a closed-loop conveyor where
et al., 2017), so that further reductions on non-value adding walk- units circulate until the complete batch has arrived in the
ing are achieved. Batching and zoning can be defined as follows loop (Johnson, 1998; Meller, 1997). A loop conveyor for a
(also see de Koster et al., 2007): huge number of orders, however, requires excessive space
on the shop floor. Therefore, especially large-sized facilities
• Batching: Instead of returning to the central depot each time a rather apply automated storage and retrieval systems (ASRS),
single picking order is completed, multiple orders are unified to e.g., crane-operated high-bay racks (Boysen & Stephan, 2016)
a batch of orders jointly assembled in a picker tour. Only if the or carousel systems (Litvak & Vlasiou, 2010), to temporar-
complete batch of orders is assembled the picker returns to the ily store bins in a more space-efficient manner. Addition-
depot and starts the next tour with the successive batch. In this ally, these storage systems allow to release bins in a specific
 N. Boysen et al. / European Journal of Operational Research 277 (2019) 396–411 401

Fig. 3. Put-to-light rack (Source: Lightning Pick Technologies, left) sliding shoe sorter (Source: Vanderlande, right).

order toward the consolidation area, so that orders are not subsequent customer order. Automated sortation conveyors
spread over an excessive number of conveyor segments. Or- revert the assessment of put walls. They reduce wage costs,
ders kept closer together on the belt reduce the time span but require a higher investment and the fixed hardware is
consolidation capacity is blocked by the collection of orders. less scalable.
(c) Finally, the bins of a batch reach the consolidation area. Here,
individual picking orders are to be assembled, packed into In the layout design phase a decision on the number (and size)
their shipping cartons and forwarded to the truck trailers of zones as well as a suited intermediate storage system and the
serving the respective destinations. There exist two alter- right consolidation technique has to be made. Due to the long tra-
native solutions of how the consolidation process is orga- dition of batching and zoning, a large body of literature has ac-
nized. A manual consolidation applies so-called put walls cumulated with regard to the first decision. The survey papers of
(see Fig. 3 (left)). A put wall is a simple reach-through rack Gu et al. (2007) and de Koster et al. (2007) summarize the re-
separated into multiple shelves, which are accessible from search efforts on how to partition the total warehouse into an ap-
both sides. Each shelf is temporarily assigned to a separate propriate number of zones. After the publication of these survey
order. On one side of the wall, a logistics worker (for a better papers, quite a few new articles on the topic of sizing zones in an
distinction called the putter) scans a unit taken from the cur- e-commerce environment has been published (i.e., de Koster, Le-
rent bin and a put-to-light mechanism indicates into which Duc, & Zaerpour, 2012; Melacini, Perotti, & Tumino, 2011; Parikh &
shelf it is to be put. In this way, bin after bin is sorted into Meller, 2008; Yu & de Koster, 2009), which are summarized in the
the wall. On the other side of the wall reside other logis- following. Yu and de Koster (2009) and Melacini et al. (2011) study
tic workers (called the packers). Here, another put-to-light progressive zoning, which due to the delays of the successive zone
mechanism indicates completed orders, so that a packer can visits seems not well suited for the tight deadlines of e-commerce
empty an indicated shelf and pack the respective units into (see above). Parikh and Meller (2008) introduce a cost model sup-
a cardboard box. Packed orders are, finally, handed over to porting the choice between a batch and a zone picking strategy.
another conveyor system bringing them toward the ship- In modern e-commerce warehouses, however, both strategies are
ping area. Put walls are, for instance, applied in the Poznan usually applied in parallel. de Koster et al. (2012) investigate the
(Poland) facility of Amazon (Boysen et al., 2018c). The big problem of determining the optimal number of zones. They intro-
advantage of a manual consolidation process is its scalability. duce a mixed-integer program to assign order lines to picking tours
By extending the put wall and/or adding additional putters minimizing the throughput time of complete orders. Having mul-
and packers consolidation capacity can quickly be adapted tiple zones and a picking capacity necessitating multiple picking
to an increasing workload. On the negative side, there are tours per zone and batch, the assignment of units to picking tours
high labor costs, so that other warehouses apply automated impacts the performance of the consolidation process. The authors
solutions where human putters are substituted by a sorta- aim to assign units to tours, such that both processes are synchro-
tion conveyor. The central conveyor system, first, delivers nized and the total throughput time, including picking, consolida-
the bins toward an induction station where either a human tion, and packing, is reduced. Based on this, the optimal number
worker or fully automated induction technology (see Gallien of zones is determined for the example of a Dutch online retailer.
& Weber, 2010) depletes the bins and isolates the units on To the best of the authors’ knowledge, no research on sizing zones
separate segments, e.g., of a sliding shoe sorter depicted in in mixed-shelves warehouse is available yet. A similar conclusion
Fig. 3 (right). Depending on the SKUs to be handled and the can be drawn for the choice of the consolidation technique. Only
sorting throughput to be reached, other sorter technology, Russell and Meller (2003) provide decision support for this deci-
e.g., crossbelt, bomb-bay, and tilt tray sorters (see Briskorn, sion task. They address the choice between manual and fully au-
Emde, & Boysen, 2017; Johnson & Meller, 2002), can be ap- tomated order consolidation in a warehouse with a picking area
plied. After passing a recognition station, where the bar code directly connected to the consolidation area via a conveyor. Finally,
is scanned to identify each unit, the belt moves along a se- no publication on the choice of a suited intermediate storage de-
quence of successive exit lanes. Once the lane currently as- vice, i.e., loop conveyor vs. high-bay rack vs. carousel, is available,
signed to the customer order of the present unit is reached, so that further research on all decisions of the layout design phase
the sorter is initiated and moves the unit – often via a grav- seems necessary.
ity chute – to the respective packing station assigned to the The problems occurring during storage assignment in a sin-
exit lane. As soon as all units of a customer order are col- gle zone are basically the same as for mixed-shelves warehouses,
lected, a beacon indicates the new status, a packer moves because batching and zoning are typically coupled with mixed-
to the station and fills the collected units into a cardboard shelves storage in e-commerce warehouses. For the sake of con-
box. This releases the lane, so that it can be assigned to a ciseness, we, therefore, refer to Section 3. On a higher planning
402 N. Boysen et al. / European Journal of Operational Research 277 (2019) 396–411

level, however, warehouse managers additionally have to distribute model for offline batch formation under the S-shape routing pol-
incoming goods among picking zones. Decision criteria are, for in- icy. Making use of newly developed bounds several state-of-the-art
stance, an equal fill level of inventory in all zones and the avail- heuristics are evaluated. Their results are improved by Hong and
ability of each SKU in multiple zones to increase flexibility. This Kim (2017) who also consider batch picking with S-shaped routes.
decision is tackled by Yuan, Cezik, and Graves (2018) based on a Žulj, Kramer, and Schneider (2018) propose a more general setting
flexibility perspective inspired by Jordan and Graves (1995). The where any routing policy can be employed. They propose a hybrid
authors assume an initially empty warehouse and benchmark eight of adaptive large neighborhood search and tabu search, which is
policies for distributing units among zones. Aiming at a high ser- shown to outperform all existing solution procedures, e.g., the ones
vice level when having limited picking capacity in each zone, they of Henn and Wäscher (2012), Oncan (2015), and Koch and Wäscher
identify two policies being superior. The first one tracks expected (2016).
workload in each zone and assigns incoming SKUs to zones with A straightforward alternative to explicitly integrate the basic
lower utilization. The second policy assigns half of the total stock urgency-efficiency trade-off of batch formation into offline batch-
to two randomly selected zones each. In this way, SKUs can be ing is to assign each order a due date. A due date is, typically,
picked from alternative zones, which increases the flexibility when defined by a cut-off date, which leaves enough time after pick-
having to react to varying workloads or picker workforces in dif- ing to pack the order into a cardboard box and to timely reach
ferent zones. the scheduled departure of the dedicated truck trailer. In this con-
The outstanding operational problem in a batching, zoning, and text, Henn and Schmid (2013) aim to minimize the total tardiness
sorting context is the batch formation problem. Batch formation, of all orders and propose an iterated local search procedure as
i.e., the decision which picking orders should be jointly processed well as an attribute-based hill climber approach for solving the re-
per picker tour, is a very active field of research and plenty rele- sulting problem. In a comprehensive computational study, the ap-
vant papers have accumulated over the past decades. An overview proaches are benchmarked for different warehouse settings and it-
is, again, provided by the in-depth survey papers of Gu et al. erated local search shows superior. In the setting of Zhang, Wang,
(2007) and de Koster et al. (2007). We only review the papers pub- and Huang (2016), vehicles’ departure times have to be met. In a
lished after these surveys, which is still a considerable amount. first setting, they minimize the sum of service times of batches.
The basic trade-off to be resolved during batch formation is the In a second setting, they maximize the quantity of orders timely
one between urgency and efficiency. On the one hand, urgent or- reaching their trucks. They, additionally, compare the results of
ders, e.g., of customers taking part in premium delivery programs both objectives and adapt existing solution procedures to the new
and with nearing promised delivery time, should be preferred and requirements. The currently best solution approach to the prob-
added to the next batches. On the other hand, orders consisting of lem setting minimizing total tardiness is provided by Menéndez,
units with close-by storage positions should be unified to batches, Bustillo, Pardo, and Duarte (2017a). They develop a variable large
such that the resulting picker tours are shortened. neighborhood search procedure and benchmark it against state-of-
The most basic approach to consider this trade-off is implic- the-art solution procedures. Based on the former work of Chen,
itly contained in offline order batching. Here, it is assumed that Cheng, Chen, and Chan (2015) and Scholz, Schubert, and Wäscher
the set of most urgent orders has already been selected (accord- (2017) study a holistic planning approach for batch formation, as-
ing to some urgency criterion outlined above). The result is a de- signing batches to pickers, sequencing of batches, and picker rout-
terministic set of (equally urgent) picking orders, which can be ing. To optimize large instances, a variable neighborhood descent
partitioned into batches merely according to the maxim of find- algorithm is suggested. In comparison to considering all subprob-
ing efficient picker tours. This branch of batching research has lems separately, the authors show that a holistic approach is able
a comparatively long tradition and current research mainly ad- to reduce total tardiness by up to 84%. Tsai, Liou, and Huang
dresses the development of high-performance solution procedures. (2008) treat an alternative problem setting where orders face ear-
Ho and Tseng (2006) as well as Ho, Su, and Shi (2008) investigate liness and tardiness costs.
seed algorithms for classical offline batching. They combine sev- A relatively new branch in this field is online (or dynamic)
eral seed-order selection rules with other selection rules for ex- order batching. Instead of partitioning a previously known, de-
tending picker tours. These combinations are tested on two differ- terministic set of orders into static batches, the dynamic arrival
ent pick-frequency distributions, i.e., derived from dedicated and of new orders is explicitly considered here. Online batching can
random storage, and for two route-planning algorithms, i.e., largest further be subdivided into fixed-time-window batching (see, e.g.,
gap (see Hall, 1993) and largest gap with simulated annealing im- Bukchin, Khmelnitsky, & Yakuel, 2012; Henn & Wäscher, 2012),
provement. Henn, Koch, Doerner, Strauss, and Wäscher (2010) in- where all new orders are considered that have arrived during a
troduce two novel solution procedures. The first approach is based fixed time span, and variable-time-window batching (see, e.g., Le-
on iterated local search and the second one on a rank-based ant al- Duc & de Koster, 2007; Xu, Liu, Li, & Dong, 2014), where batch
gorithm. Both algorithms are compared and shown to outperform formation is triggered whenever a predefined number of orders
existing procedures. Due to its shorter solution time the authors, have arrived. By applying analytical models, van Nieuwenhuyse
finally, recommend the iterative local search approach. Hsieh and and de Koster (2009) investigate the impact of varying setup pa-
Huang (2011) propose solution approaches on the basis of data rameters on both online batching strategies, e.g., variable- and
mining techniques. Hong, Johnson, and Peters (2012b) provide a fixed-time-window batching, in a 2-block warehouse with multi-
fast heuristic, based on a decomposition approach, that is suitable ple pickers, general setup and service time distributions, as well
for large instances of the offline batching problem. The authors as a downstream packing process. While it is assumed that the
limit themselves to traversal routing (see Hall, 1993) and suggest online batching strategy is given, the computational study clearly
a new MIP model for the batching problem making use of this shows that a pick-and-sort configuration is superior to the sort-
routing policy. Employing a novel, tight lower bound they show while-pick strategy in nearly every setting. This means that the
that the approach results in good quality solutions even for in- additional effort for sorting the picked units into bins while the
stances with more than 20 0 0 orders and 10 aisles within a ware- picker is still on the move during sort-while-pick order process-
house. For this instance size, the approach consumes between 60 ing is higher than the additional double handling caused by the
and 80 seconds on average. Kulak, Sahin, and Taner (2012) and Li, two successive picking and sorting process steps of a pick-and-sort
Huang, and Dai (2017) also propose a decomposition of the joint strategy. In addition to that, extensive experiments on other sys-
batching and routing problem. Bozer and Kile (2008) provide a MIP tem parameters (e.g., the warehouse size or the average customer
 N. Boysen et al. / European Journal of Operational Research 277 (2019) 396–411 403

order volume) are executed to assist practitioners during the lay- from all zones in parallel, in addition to the urgency-efficiency
out phase. Henn and Wäscher (2012) adapt solution procedures trade-off, an equal distribution of the workload among the zones
for offline batching to the online case. They consider several ex- should be considered. A fair distribution of workload ensures that
isting batching algorithms and selection rules for the batch to be all partial orders are finished at about the same time and can
processed next and compare them. In their computational study, advance jointly into the consolidation area. If some zones, how-
a combination of iterated local search and selecting the batch with ever, receive a considerably higher workload their partial batches
highest savings of processing time is identified to result in the low- may arrive late, which bears the risk that consolidation capacity
est order completion times. Xu et al. (2014) provide an analytical is blocked by orders waiting for late units. In the worst case, this
model for a variable-time-window batching environment, imple- may lead to deadlocks where all sortation lanes are blocked, newly
menting random storage and S-shaped routing. They measure the arriving orders cannot be sorted, and costly recovery procedures
effect of varying order and system parameters, e.g., the number for the steadily succeeding units need to be executed. Çeven and
and length of picking aisles, the interarrival times of orders, and Gue (2015) give advice regarding the optimal number and timing
the batch sizes, on the expected throughput time. of waves in such an environment. They develop analytical models
Another stream of batching research tries to extend offline and and test their approach on real-world data from a large distribu-
online batching by further peculiarities of practical relevance. For tion center. It is shown that an optimized wave release strategy
instance, order batching and sequencing in narrow aisles where significantly improves the ratio of shipments on time in compari-
pickers cannot pass each other is investigated by Hong, Johnson, son to a simple rule of thumb. The decision on whether wave pick-
and Peters (2012a). They consider congestion among pickers, typi- ing should be applied at all is tackled by Gallien and Weber (2010).
cally resulting in reduced pick performance (see also Gue, Meller, The alternative is to directly release each incoming order without
& Skufca, 2006; Parikh & Meller, 2010), and present an integer pro- collecting them for a while and unifying them to waves that are
gram as well as a heuristic based on simulated annealing for the jointly collected in all zones. The authors compare waveless and
resulting optimization problem. In their computational study, the wave-based picking strategies in an e-commerce warehouse, uti-
holistic optimization approach leads to a 5–15% shorter total re- lizing a newly developed queueing model. Based on the results
trieval time compared to a successive execution of batch formation gained by applying this model to real-world data, an optimiza-
and picker routing. Matusiak, de Koster, and Saarinen (2017) in- tion technique based on dynamic programming for the timely re-
vestigate a special offline batching problem where different per- lease of orders in a waveless environment is introduced. In a thor-
formance levels of pickers are considered. They solve the parti- ough simulation study, superiority (or at least equal performance)
tioning of orders into batches, the assignment of these batches of waveless picking in all considered scenarios is shown when ap-
to pickers, and the picker routing problem in a holistic approach, plying their approach. An alternative to integrate the coordination
such that the total picking time is minimized. They also propose of multiple zones into offline order batching has been introduced
an estimation method for batch execution times and an adaptive by Gademann, van den Berg, and van der Hoff (2001). They sub-
large neighborhood search heuristic for the integrated problem. divide the total set of orders into a given number of waves by
The methods are tested on data of a large retail warehouse and minimizing the maximum processing time over all zones summa-
improvements of nearly 10% of total batch execution time are ob- rized over all waves. This problem is reconsidered by Menéndez,
tained. A similar problem setting for online order batching is in- Pardo, Snchez-Oro, and Duarte (2017b). They suggest a new paral-
vestigated by Zhang, Wang, Chan, and Ruan (2017). They do not lel variable neighborhood search, which outperforms previous so-
assume different processing times of pickers, but only coordinate lution procedures.
the assignment of batches to the workforce. The same setting is Batch formation is (by far) the most well researched problem
investigated for the offline case by Henn (2015). Another special or- within the scope of our survey. However, there still remains room
der batching problem is addressed by Matusiak, de Koster, Kroon, for future research. In today’s e-commerce warehouses batching
and Saarinen (2014) where precedence constraints among picked and zoning is typically coupled with mixed-shelves storage. In a
units have to be considered. Precedence constraints may result mixed-shelves warehouse, SKUs are stored at multiple positions
from weight and stability aspects on a picking cart or different and, thus, also in different zones. Therefore, batch formation has
shapes of products. The problem is modeled and a suited heuris- additionally to decide from which zone a specific unit should be
tic solution procedure based on a decomposition approach is pre- picked. To the best of the authors’ knowledge, no publication on
sented. In a computational study, the new approach is tested on the special batching problems arising from this peculiarity is avail-
real-world data of a large Finnish warehouse. Reductions of the able yet. This finding is also valid for some further operational
total travel distances about 15% compared to the currently imple- problems beyond batch formation that are addressed in the follow-
mented routing policy are recorded. Grosse, Glock, and Ballester- ing.
Ripoll (2014) consider weights restrictions directly in their joint The release sequence of bins from intermediate storage contain-
order batching and picker routing approach. Wruck et al. (2013) in- ing the order batches to be sorted is considered by Boysen, Fedtke,
tegrate forward and return product flows, caused by customer re- and Weidinger (2018a) and Boysen et al. (2018c) for automated
turns, into one single approach and solve the combined batching and manual sorting systems, respectively. Boysen et al. (2018a) de-
problem of timely delivering customers and efficiently stowing re- scribe the minimum order spread sequencing problem, which aims
turned products. Suited solution procedures, based on seed algo- to release the bins in such a way that the number of sorter seg-
rithms, are proposed and tested on real-world data of a library ments between the first and the last unit of an order on a sortation
warehouse. The combined approach leads to up to 44% shorter to- conveyor is minimized. In this way, orders can quickly be assem-
tal travel distances in comparison to handling picking and stowing bled in their dedicated packing stations and consolidation capac-
separately. ity of the automated sorter can be released earlier. The problem
Batch formation is even more complicated if, in addition to is formalized, NP-hardness is proven, and a suited solution proce-
batching, the warehouse is subdivided into multiple zones. In this dure based on dynamic programming is introduced. In a simula-
case, the wave of orders to be processed next determines the work- tion study, superiority over randomized bin release sequences is
load in all zones and leads to an interdependent batching problem shown and the impact of multiple additional problem parameters
in each zone. The parts of the wave of orders to be picked in the on order consolidation performance is investigated. Among them
respective zone, additionally, have to be partitioned into batches in is, for instance, the impact of the number of bins a given batch is
each zone. When selecting the wave of orders to be picked next divided into. Having a manual consolidation process, the resulting
404 N. Boysen et al. / European Journal of Operational Research 277 (2019) 396–411

problem is described by Boysen et al. (2018c). They consider the dynamic order processing is applied, but rather depends on suffi-
release of bins from intermediate storage toward a manual put cient sortation capacity for isolating the orders.
wall. They model the problem as a special machine scheduling Dynamic order processing is an organizational adaption of
problem with a m:n-relationship among jobs (bins). The problem is picker-to-parts systems, so that, during the layout design phase and
proven to be NP-hard and exact and heuristic solution procedures storage assignment, the same general decisions as elaborated in the
are suggested. A significant reduction on both, idle times of pack- previous section have to be taken (see Section 4). A further in-
ers and throughput time of batches, is shown in a simulation study teresting long-term research task is the question under which cir-
for optimized bin sequences. Similar to Bozer, Quiroz, and Sharp cumstances either milk-run or interventionist order picking should
(1988), Meller (1997) describes a planning problem and proposes be preferred. A main influencing factor is certainly the size of the
solution procedures for assigning exit lanes and packing stations assortment to be handled. Calculating the break-even point up to
to orders in an automated sorting system. Both papers consider no which assortment size the prolonged routes of milk-run picking are
intermediate storage system but a direct release from the picking acceptable and still outweigh the larger turmoil of continuous tour
area into the consolidation system. They propose an optimization replanning required by interventionist picking is a challenging task
procedure for the order-to-lane assignment given a fixed sequence for future research.
of units on the sortation conveyor. In a computational study, a Short term operational problems include the update mechanism
reduction of average sortation time by more than 40% is shown during the tour. Gong and de Koster (2008) study milk-run picking
when applying the optimization procedure in comparison to a sim- by applying stochastic polling theory (Srinivasan 1991). They sur-
ple decision rule applied in business practice. Johnson (1998) in- vey three different update mechanisms for pick lists, finding that
troduces a stochastic model for the same setting and shows that, their so-called exhaustive service policy performs best. Under this
given little lane blockings, assigning the next order to a free lane policy, all requests of a SKU are processed, even if the request is re-
outperforms more complex rules. ceived during processing that very SKU. It is shown that dynamic
order processing outperforms traditional wave-based batch picking
systems in an environment without zones and a single order line
5. Dynamic order processing per customer order. They, furthermore, provide decision support
on the workforce size depending on the cost structure of workers
In traditional order processing, a pick list is fixed once or- and disappointed customers as well as depending on a maximum
der picking is initiated. This policy results in distinct waves each waiting time of orders. Based on this, van der Gaast, de Koster,
picked, sorted, and shipped one after another. Dynamic order pro- and Adan (2018) benchmark a basic batch picking approach with
cessing systems, however, allow pick list updates even after the milk-run picking with single and multi-line orders picked from a
respective picking tour has started. Incoming urgent orders are in- zoned warehouse. In line with the previous paper, order through-
stantaneously added to the current picker route and there is no put time as well as waiting time of orders can be improved when
need to waitlist them until the next wave is processed. Whenever applying dynamic order processing. Lu, McFarlane, Giannikas, and
a new order arrives, it has to be decided whether or not this order Zhang (2016) address interventionist picking. They determine a
should be dynamically added to the current tour of a picker. Such (new) shortest picker tour whenever the pick list of a picker is up-
a decision is, for instance, required if the picking capacity, e.g., on dated and, thus, integrate detours into the current picker route in
a picking cart, is scarce and adding the new order necessitates a order to access new units. They introduce a novel routing algo-
postponement of earlier orders already on the tour. rithm for this purpose, based on the one of Ratliff and Rosenthal
Furthermore, adding a new order may require to alter the cur- (1983). In their simulation study, they benchmark their approach
rent picker tour. No tour adaption is required if each picker always with static batching and show that the average completion time of
traverses his/her complete zone on any tour anyway. Then, new or- orders is significantly reduced. These performance gains, however,
ders only need to be indicated by a display or a light signal at the induce longer travel distances of pickers. Giannikas, Lu, Robertson,
respective shelves and the picker collects these units when passing and McFarlane (2017) propose another interventionist order pick-
by. As pickers operate on fixed tours in such a system, this mode ing strategy. In their paper, they compare several combined strate-
of dynamic order processing is also denoted as milk-run picking. If gies for dynamic order batching and interventionist pick list up-
pickers do not apply fixed tours through the complete warehouse, dates and compare them to a purely static and a purely dynamic
but specifically address the current pick list on individual tours, batching strategy. Considering the performance indicators average
then adding a new order requires an instantaneous adaption of the order completion time and average travel distance, it is shown that
current picker tour. This mode of dynamic order picking is often the new approach significantly outperforms both benchmarks.
called interventionist order picking. Once additional units arrive, the
current tour is discarded and a novel tour including the not yet 6. AGV-assisted picking
picked and novel units is quickly generated and announced to the
picker, e.g., via a handheld or pick-by-voice system. In both kinds Another alternative to reduce unproductive walking times in a
of dynamic order processing systems, picked units have to be con- picker-to-parts system is to support the pickers with automated
solidated after picking into distinct orders, so that dynamic order guided vehicles (AGVs). These AGVs carry the picked units in bins
processing requires sortation technology (see Section 4) (Gong & (or on pallets, containers, or roll cages if large-sized SKUs are
de Koster, 2008). processed) and autonomously accompany pickers on their way
The biggest advantage of dynamic order processing is certainly through the warehouse. Each picker puts his/her units onto the
the flexibility to quickly process urgent orders, which is especially AGV and once the current orders are complete, the AGV au-
valuable for the tight schedules of online retailers. This flexibility tonomously returns to the depot, while the picker remains in the
comes at the price of a frequent replanning of picker tours if in- storage area. A new AGV is requested to meet the picker at the
terventionist picking is applied. Milk-run picking struggles if large first storage position of the successive pick list. In this way, pickers
assortments have to be handled. Then, the fixed tours through the can continuously pick order after order without intermediate re-
complete warehouse may become fairly large compared to individ- turns to a depot. We call this the fixed-assignment policy. In an al-
ual tours just addressing the current subset of units to be actually ternative setting, AGVs are not fixedly assigned to a specific picker
picked. The other requirements of online retailers, i.e., the ability during processing the current order. They autonomously drive to-
to process small orders and varying workloads, is not affected if ward a picking position and wait there until some picker loads the
 N. Boysen et al. / European Journal of Operational Research 277 (2019) 396–411 405

lar warehouse and integrate it into a dynamic programming ap-

proach, so that for a given order sequence the minimum makespan
can efficiently be found in polynomial time. This procedure is in-
tegrated into myopic search procedures and a genetic algorithm to
also evaluate alternative order sequences. Future research should
address the alternative picking policy of free-floating pickers. This
policy requires an interdependent scheduling of all AGVs and all
pickers under synchronization constraints (i.e., for picking an AGV
and a picker must be simultaneously present at the respective
picking location). Also, short-term reactions to unforeseen delays,
which require a quick re-routing of AGVs and/or pickers, is a chal-
lenging task for future research.

7. Shelf-moving robots

Mobile robot fulfillment systems (MRFS) are parts-to-picker sys-

tems where mobile robots are able to lift movable racks (also de-
noted as inventory pods) and bring them directly toward station-
ary pickers operating in workstations. According to Azadeh et al.
Fig. 4. AGV for small-sized items with trailer.
(2018), Jünemann (1989) was the first to conceptualize MRFS. Af-
terwards, the system was brought to the market in 2003 (Guizzo,
2008) and U.S. patented in 2008 (Mountz et al., 2008) by KIVA Sys-
requested units, then they move to the next position where an-
other picker can execute the pick (Azadeh et al., 2018). We call this tems Inc., which is why MRFS are widely known under the name
KIVA systems and the mobile robots as KIVA robots. In 2012, Ama-
mode of operation the free-floating policy.
zon acquired the company, renamed it to Amazon Robotics, and,
AGV-assisted order picking is often applied for heavy and bulky
items. Products such as white goods, large consumer electronics or now, applies the system in many of its U.S. distribution centers.
Since then, other providers have entered the market with compa-
carpets cannot be carried by the picker all the way back to the de-
rable mobile robots (Banker, 2016; Kirks, Stenzel, Kamagaew, & ten
pot, so that vehicle support is required anyway. The additional in-
Hompel, 2012) – one of them, CarryPickTM of Swisslog, is depicted
vestment for upgrading a traditional forklift to an AGV is compara-
tively low. However, AGV-assisted picking is not bound to heavy in Fig. 5 (left) – and other online retailers such as Alibaba apply
MRFS too (Techinasia, 2017).
and bulky goods; Fig. 4 depicts an AGV for small-sized items4 .
In the KIVA system, robots have a rotation and a lifting mech-
Small-sized AGVs can also be applied to support pickers in mixed-
shelves warehouses (see Section 3) and in a zoning and batching anism and are electrically powered. For their orientation, the shop
floor is subdivided into a grid and each square is labeled with a
environment (see Section 4), so that they are a valid alternative for
barcode. The robots’ integrated camera system continuously reads
online retailers too.
Layout design phase: The most important long-term decision these barcodes to locate themselves and the rotation mechanism
allows them to move rectilinearly from square to square. The lift-
when setting up an AGV-assisted order processing is certainly the
ing unit is able to lift more than 1,0 0 0 kilograms so that a robot
interdependent sizing of AGV fleet and picker workforce. A bot-
tleneck on either side is to be avoided, because otherwise idle can drive directly under a man-high rack, lift it, and bring it from
the storage area to a stationary picker (D’Andrea & Wurman, 2008).
times of either pickers or AGVs will considerable reduce picking
In a picking station (see Fig. 5 (right)), multiple customer orders
performance. Scientific decision support on this problem is yet not
available. When closing this gap, future research should also con- are processed concurrently. Assisted by a pick-to-light system, e.g.,
sider the impact of the selected picking policy (fixed-assignment a laser pointer (Wurman, D’Andrea, & Mountz, 2008), the picker
retrieves units from the current rack and puts them into bins as-
vs. free-floating) on this decision. The free-floating policy promises
sociated with the active customer orders. In a continuous process,
a smaller picker workforce, but is prone to short-term plan alter-
ations due to picker delays. racks are successively moved toward a station, units are retrieved,
customer orders are completed and replaced by new bins of sub-
Existing storage assignment policies for traditional picker-to-
sequent orders. The system is described in even more detail by
parts warehouses, such as full turnover storage and class-based
storage, see de Koster et al. (2007), aim to store frequently re- D’Andrea and Wurman (2008), Wurman et al. (2008), and Enright
and Wurman (2011).
quested SKUs closer to the depot. With AGV assistance, however,
Due to the elimination of non-value adding picker walking
pickers do not have to return to the depot, but remain in the ware-
house, so that fast-moving items should rather be stored some- MRFS have a high picking performance, which is reported to
reach 600 and more order lines per hour and picker (Wulfraat,
where in the center of the warehouse or in areas easily accessible
2012). This makes them well suited for the tight deadlines of e-
via (middle) aisles. An adaptation of traditional storage assignment
rules and a development of new rules for AGV-assisted picking is commerce. Also, small order sizes and a large assortment seem
an interesting field for future research. unproblematic, as long as there is enough space for the inventory
pods on the shop floor. In contrast to other parts-to-picker sys-
Order picking with AGV assistance has only been considered by
tems, MRFS have a flexible layout and come by without fixedly in-
Löffler, Boysen, Glock, and Schneider (2017) yet. They optimize the
routing of a single picker under the fixed-assignment policy. For a stalled hardware. Thus, by adding pods and robots to or removing
them from a facility MRFS are easily scalable to varying workloads.
given set of orders, they aim to minimize the makespan of pick-
During the layout design phase mainly the following decisions
ing, if AGVs are no bottleneck. They modify the efficient algorithm
of Ratliff and Rosenthal (1983) for picker routing in a rectangu- have to be made: How many picking and replenishment sta-
tions for picking orders and refilling the mobile racks, respectively,
should be erected and where should these stations be located?
4
The picture is published under the Creative Commons Attribution ShareAlike The latter question is closely interdependent with the sizing and
License. The author of the picture is AGVExpertJS. the layout of the storage area for the inventory pods. Existing
406 N. Boysen et al. / European Journal of Operational Research 277 (2019) 396–411

Fig. 5. CarryPickTM (Source: Swisslog, left) and picking station (Source: Amazon Robotics, right).

MRFS storage areas copy traditional warehouse layouts with par- is the one of Weidinger, Boysen, and Briskorn (2018). They aim to
allel aisles dedicated to robot movement and accessing the racks find an assignment of pods to parking positions, which minimizes
stored in parallel lines along the aisles. Empty robots, however, the travel distance of robots loaded with racks to satisfy a given
can drive unobstructed below the racks, so that also pods with- set of rack visits at different picking stations. A mathematical def-
out direct access to an aisle can easily be reached. Other robots inition of the problem, a proof of NP-hardness even for a rectan-
could then be applied to dynamically free an aisle, so that compact gular storage layout and a suited matheuristic are provided. In a
storage layouts similar to puzzle-based storage systems (see Gue, simulation study, the approach is compared to five other, simple
Furmans, Seibold, & Uludağ, 2014; Gue & Kim, 2007) could eas- storage assignment rules, e.g., random storage. It is shown that the
ily be realized. The only paper addressing layout aspects of MRFS size of the robot fleet as well as the total travel distance of the
is provided by Lamballais, Roy, and de Koster (2017a). They pro- robots is significantly reduced when optimizing the parking posi-
vide an extensive performance analysis of different (traditional) tions of racks. However, one of the simple storage assignment rules
layouts employing queuing models, which are validated by a sim- (denoted shortest-path storage) is shown to deliver nearly as good
ulation study. They focus their study on a single workstation and solutions as the sophisticated optimization approach.
transfer their findings to a full warehouse. In consequence, robots Short-term order picking has to select a specific pod to satisfy
are assigned exclusively to a single workstation. The study shows the demand of an order that has to be fulfilled at a specific pick-
that the performance measures considered, i.e., robot utilization, ing station. Once these decisions are made, a specific robot is to
the mean length of external order queues, the average order cy- be assigned to the resulting movement of the pod, which also in-
cle time, and the utilization of the workstations, are robust to the volves finding a suited travel path and coordinating it with com-
length-to-width ratio of the warehouse but highly sensitive to the peting robots. To the best of the authors’ knowledge, no litera-
location of the workstations. Furthermore, they evaluate the im- ture exists regarding the assignment of orders to picking stations.
pact of the storage assignment policy, i.e., they compare random Instead, existing research assumes that an assignment of orders
storage with a zone-based assignment where fast-moving items to picking stations has already been determined. Under these cir-
are stored closer to the stations. cumstances, Zou, Gong, Xu, and Yuan (2017) tackle the assignment
Storage assignment in MRFS can be subdivided into two basic of robots to picking stations having different picking performance
decision tasks. On the one hand, incoming units referring to differ- due to individual processing speeds of the respective pickers. For
ent SKUs have to be put away into specific inventory pods. Existing the solution of the resulting optimization problem they provide
research (Boysen et al., 2018b; Lamballais, Roy, & de Koster, 2017b) a neighborhood-based optimization approach. Solutions are eval-
has shown that applying mixed-shelves storage where units of the uated via a semi-open queueing network, which itself is verified
same SKU are spread over multiple racks greatly improves picking by a simulation study. It is shown that the neighborhood search
performance. In this case, multiple alternative pods are available is able to find near optimal solutions and optimized assignments
for satisfying a specific customer demand, so that the probabil- are superior to randomized ones. In the paper of Yuan and Gong
ity is high that a suited rack stored close to the respective pick- (2017), a dedicated assignment of robots to picking stations is com-
ing station is available. This reduces the probability of idle pickers pared with a pooled approach. The latter means that robots can
waiting for pods that have not yet arrived and reduces the travel deliver pods to all stations and are not reserved for single pick sta-
distances of the robots. Once the composition of the racks is de- tions. It is shown that the throughput time can be considerably re-
termined, the racks themselves need to be put into storage. After duced when using the pooled strategy. The study is based on open
picking, robots need not take pods back to their previous storage queuing networks, which are also employed to determine the op-
position in the storage area. Instead, any other open storage posi- timal number and velocity of robots as well as the ratio of robots
tion can be selected, e.g., closer to the picking station to be vis- to pickers. The assignment of racks to orders of a station is tackled
ited next, so that a dynamic storage assignment, which rather re- by Boysen et al. (2018b). They consider a single picking station and
sembles a short-term parking problem, becomes relevant. In this aim at a schedule for a given set of orders. Given that a predefined
context, Lamballais et al. (2017b) study the impact of the number number of orders can be picked in parallel on the workbench of
of pods storing the same SKU, the ratio of picking stations to re- the picking station, they search for a sequence of order batches and
plenishment stations, and the replenishment level of the pods on rack visits minimizing the total quantity of racks to be transported
the mean throughput time of an order. They confirm the result of to the station in order to satisfy all orders. A suited decomposition
Boysen et al. (2018b) that a higher degree of scatteredness of units approach is presented and tested in a simulation study. According
within pods reduces the mean processing time of orders. Addition- to this study, the size of the robot fleet can be more than halved
ally, a ratio of twice as much picking as replenishment stations and when using an optimized schedule. Additionally, it is shown that
a replenishment level of 50% is identified as a promising configu- the more scattered the SKUs among the racks, the fewer rack visits
ration. The only existing paper on the short-term parking problem are required to satisfy a given set of orders. Once the movements
 N. Boysen et al. / European Journal of Operational Research 277 (2019) 396–411 407

of inventory pods between picking stations and storage area have

been derived, the routing of the robots becomes relevant. A vast
body of literature is available on routing AGVs, which is summa-
rized, e.g., in the survey papers of Qiu et al. (2002) and Vis (2006).
We only review the literature published since then and suited to
MRFS. Herrero-Pérez and Martínez-Barberá (2011) decompose the
coordination of multiple AGVs into path planning, obstacle avoid-
ance and traffic control which are solved in a decentralized man-
ner. They present suited solution procedures, which are tested in a
simulation study as well as a real-world scenario. Other papers are
explicitly treating MRFS. Yu (2016) studies path planning of mul-
tiple robots on a planar graph. NP-hardness of four basic problem
settings is proven and sharing paths among robots with opposing
directions is identified to considerably complicate the problem. The
idea of having an adaptive highway system has been brought up
by Roozbehani and D’Andrea (2011). They try to find one-way ex-
pressways in real time to maximize the average speed of moving
robots.

8. Advanced picking workstations

Advanced picking workstations (also denoted as picking bays,

see Dallari, Marchet, & Melacini, 2009) promise a picking per-
formance of up to 10 0 0 order lines per hour (de Koster, 2008;
de Koster et al., 2007). They resemble the picking stations of the
KIVA system, but are ergonomically designed and automatically fed
with bins from a directly connected storage system via conveyors.
An order fulfillment system based on picking workstations consists
of the following three elements:

• Storage system: The bins containing the SKUs are stored in an

automated storage system. Typical systems are either one or
multiple carousel racks (see Litvak & Vlasiou, 2010), a crane-
operated ASRS (i.e., a miniload system, see Manzini, Gamberi, &
Regattieri, 2006; Roodbergen & Vis, 2009), or a lift and shuttle
system where horizontal and vertical movement is separated
(see Azadeh et al., 2018). Fig. 6. Pick-to-tote stationTM SSI Schäfer.
• An intermediate conveyor system that delivers requested bins
between storage systems and stations. This system also buffers
bins until all bins required for a specific customer order have Andriansyah, Etman, Adan, and Rooda (2014), a picking worksta-
arrived. tion is supplied from a crane-operated ASRS. They evaluate differ-
• Finally, within each advanced picking workstation requested ent retrieval policies of storage bins from the rack and quantify
items are withdrawn by a human picker from the arriving stor- the impact of these policies on the picking performance via sim-
age bins and put into other bins (denoted as order bins) that ulation. Additionally, they propose a new design for the conveyor
are associated with customer orders (see Fig. 6). Typically, a system, a carousel mechanism to avoid deadlocks between com-
workstation has enough space on the workbench for multiple pleted and new storage bins. Further simulation studies predicting
order bins processed in parallel. A monitor assists the identi- the picking performance if a workstation is supplied from a crane-
fication of how many units to put into which order bins. One operated ASRS are provided by Andriansyah, Etman, and Rooda
after another, storage bins arrive at the station, are processed (2010) and Manzini et al. (2006). Andriansyah et al. (2010) study
by the picker until an order bin is completed and automatically a picking workstation that assembles one order at a time. The sta-
swapped with a new (empty) order bin. In this way, a parts-to- tion is connected to an ASRS via a multi-lane buffer. They aggre-
picker system is realized where the picker is relieved from any gate several effects on the pick performance in a single relevant
unproductive work and only has to pick units. performance measure called Effective Processing Time (EPT). For
the same basic setup of Andriansyah et al. (2010) the paper of
Due to their high performance advanced picking worksta- Claeys, Adan, and Boxma (2016) provides queuing models for per-
tions seem well suited for the tight deadlines of e-commerce. formance analysis. They also restrict their research to the assem-
Large assortments and small-sized orders also seem unproblematic bly of one order at a time, whereas parallel processing of mul-
for these parts-to-picker systems. However, adding new stations tiple parallel order is quite common in business practice (Füßler
and/or larger storage systems (or removing them) requires plenty & Boysen, 2017b). They study the effect of an arrival process of
constructional effort and is, therefore, barely possible on short no- bins, which is not Poisson distributed. The latter results from the
tice. The major disadvantage when applying picking workstations application of specific batching and/or storage assignment strate-
for online retailing is, thus, their lack of scalability. gies. Tappia, Roy, Melacini, and de Koster (2018) compare the in-
Research focusing on the layout design phase of picking work- tegration of advanced picking workstations with different storage
stations is scarce. The few existing approaches all address perfor- devices, i.e., crane-operated vs. lift and shuttle ASRS. By applying
mance estimation with simulation studies or queuing models in semi-open queuing models they conclude that shuttle-based ASRS
order to quickly evaluate different layout setups. In the study of yield investment cost savings (i.e., fewer aisles in the storage area
408 N. Boysen et al. / European Journal of Operational Research 277 (2019) 396–411

Table 1
Surveyed literature sorted by warehousing system and decision problem.

Layout design Storage assignment Order picking

Mixed-shelves storage 0 1 3
Batching, zoning, and sorting 5 1 42
Dynamic order processing 0 0 4
AGV-assisted picking 0 0 1
Shelf-moving robots 1 2 6
Advanced picking workstations 5 0 1

and fewer picking stations), paired with a lower total throughput

time at a given order arrival rate.
Storage assignment and order picking: Existing research in this
area rather focuses on the storage systems (in isolation). Where
to store items in an automated storage system and how to effi-
ciently schedule the retrieval and storage requests are among the
classic research questions of warehousing research. Instead of try-
ing to summarize the vast body of literature on these topics we
only refer to the most recent existing survey papers on carousel
systems (Litvak & Vlasiou, 2010), crane-operated ASRS (Boysen &
Stephan, 2016; Manzini et al., 2006; Roodbergen & Vis, 2009), and
lift and shuttle systems (Azadeh et al., 2018) instead. Literature
on the operational decision problems of advanced picking work-
stations and their interplay with the storage and conveyor system,
however, is barely existent. The only paper in this direction is pro-
vided by Füßler and Boysen (2017b). They aim to synchronize the
arrival sequence of storage bins with the batches of orders simul-
taneously processed on the workbench of a single picking work-
Fig. 7. Bag sorter system (Source: Dürkopp).
station. If orders are unified to batches, such that multiple active
orders require the current SKUs provided in a storage bin, then
fewer bins need to be delivered to the station. This relieves the
of storage, automatically sorted into the right sequence via a sys-
storage system in an indirect manner. In their paper, Füßler and
tem of switches, and conveyed to packing stations. Here, units ar-
Boysen (2017b) show that this indirect effect can indeed lead to a
rive in the proper sequence, so that one order after the other can
much larger relief of the storage system than optimizing its stor-
be packed into its shipping carton. The capacity of a bag sorter is
age and retrieval sequence directly. Future research should follow
typically not large enough to store all units of a warehouse. This
this research direction and investigate holistic systems consisting
leads to the interesting operational research question, how many
of picking workstations, conveyor and storage system rather than
units of each SKU should be put into bags facing the highly volatile
only addressing each element in isolation.
demands of customer households. During the layout design phase,
the sizing of the storage area and the required number of bags for
9. Future research needs and conclusions a smooth picking process are, thus, important issues.
System selection: A question of utmost importance for ware-
The surveyed literature for each of the reviewed systems is house managers is the question for the right system best suited for
summarized in Table 1. This summary reveals that each single their specific situation. Our discussion in this survey regarding the
warehousing system still has plenty demand for future research; suitability of each system for the special needs of online retailing is
some systems, i.e., mixed-shelves storage, dynamic order process- certainly not sufficient to guide a real-world system selection. On
ing, and AGV-assisted picking, are barely addressed at all. Only the one hand, empirical research could record the systems selected
batch formation in a batching, zoning, and sorting environment has for real-world operations depending, e.g., on the branches of indus-
received plenty attention so far. However, even for this problem try, customer and product characteristics, size and throughput of
the peculiarities of e-commerce (i.e., a widespread coupling with the facilities. First steps in this direction are provided by Marchet,
mixed-shelves storage) have not exhaustively been addressed. Be- Melacini, and Perotti (2015) and Davarzani and Norrman (2015).
yond these specific systems, we see some emerging general trends The former record the selected warehousing systems of 40 Italian
in (e-commerce) warehousing that require further scientific con- distribution centers and derive some rules with regard to the fit
sideration. of different systems. The latter compare the main research issues
Novel systems: Additional research is required whenever novel of academia obtained by a database-driven literature search with
systems are presented in order to evaluate their potential picking the needs of practitioners obtained by interviews. Davarzani and
performance and their suitability for e-commerce. One of these Norrman (2015) find a remarkable gap. Obtained empirical data
novel ideas for which implementations in large real-world facili- could also be applied within data-driven techniques such as data
ties of online retailers already exist (but no scientific literature) is envelopment analysis (DEA) to benchmark different systems (Chen,
the so-called bag sorter system (see Fig. 7). Bag sorters adapt the Gong, de Koster, & van Nunen, 2010). In this way, some systematic
basic principle of hanging goods handling and move units in bags empirical information on what warehouse systems have been cho-
hanging from trolley conveyors. The Zalando distribution center in sen in what situation could be obtained. On the other hand, simu-
Mönchengladbach (Germany), for instance, applies a large system lation studies evaluating the picking performance of a given set of
with a capacity for 50 0,0 0 0 bags servicing 80 packing stations (RP customer orders in different warehousing systems would be a la-
Online, 2016). First, individual units are filled each in a bag, which borious, yet valuable contribution. Most online retailers we visited,
are moved into the storage area. Upon request, bags are moved out however, do not apply just a single warehousing system for the
 N. Boysen et al. / European Journal of Operational Research 277 (2019) 396–411 409

complete range of products, but apply multiple systems in parallel return rates of more than 80% for single SKUs. Some online retail-
and link them via conveyors. In this way, fast-moving items can, ers have completely outsourced the management of returns. In this
for instance, be picked from mixed-shelves and seldom requested case, both business processes, order fulfillment and return man-
C products via advanced picking workstations. Decision support agement, can be optimized in a decoupled way. Whenever returns
under which circumstances multiple parallel warehousing systems are handled at the warehouses, though, interdependencies occur
should be applied in what combination is yet missing. (de Koster, de Brito, & van de Vendel, 2002). Some retailers, for
Ergonomics: Probably the most important selection criterion for example, implement the policy that all returns of a SKU have to
warehousing systems in e-commerce is picking performance. How- be shipped before factory-fresh items are picked. This leads to ad-
ever, if any non-value adding work content is removed and pickers ditional constraints for the order picking process. Additionally, re-
have to process up to 10 0 0 order lines per hour (see Section 8), turned products often have a rather high probability to be ordered
this puts excessive physical and psychological stress on the work- once again in the near future. Therefore, they could be assigned
force (Otto, Boysen, Scholl, & Walter, 2017). Given the aging some privileged storage positions. Systematic decision support on
workforce in many industrialized countries, ergonomic aspects in how to best reintegrate returns in the process steps of the forward
workplace design and a suited cooperation between automated chain of e-commerce warehouses is rare (see de Koster et al., 2002)
solutions and human workers should be considered. The survey and constitutes a challenging field for future research.
papers of Grosse, Glock, Jaber, and Neumann (2015) and Grosse, Seeing the ever growing market shares of online retailing
Glock, and Neumann (2017) on human factors in warehousing, (Statista, 2017) and the manifold unanswered questions elaborated
however, reveal that not much research in this direction exists. in this paper it seems fair to predict that warehousing will remain
Specifically, planning procedures for routing the pickers in AGV- a fruitful field of research in the years to come.
assisted order picking (see Section 6) or for dividing a wave of or-
ders among multiple advanced picking workstations (see Section 8) References
could, for instance, try to distribute the inevitable idle time where
Agatz, N. A., Fleischmann, M., & van Nunen, J. A. (2008). E-fulfillment and mul-
humans wait for machines fairly among the workforce.
ti-channel distribution a review. European Journal of Operational Research, 187,
Holistic research: To enable a rigorous analysis, most existing 339–356.
papers concentrate on an isolated problem of a single subsystem. Andriansyah, R., Etman, L. F. P., Adan, I. J., & Rooda, J. E. (2014). Design and analysis
of an automated order-picking workstation. Journal of Simulation, 8, 151–163.
Today’s warehouses, however, are large facilities where plenty re-
Andriansyah, R., Etman, L. F. P., & Rooda, J. E. (2010). Flow time prediction for a
sources and processes cooperate. Thus, from the warehouse man- single-server order picking workstation using aggregate process times. Interna-
ager’s perspective, holistic models are required that take a rather tional Journal on Advances in Systems and Measurements, 3, 35–47.
unifying look at the complete order fulfillment process. In mobile Azadeh, K., de Koster, M. B. M., & Roy, D. (2018). Robotized warehouse systems:
Developments and research opportunities. Transportation Science. In press.
robot fulfillment systems (MRFS), for instance, the assignment of Baker, P., & Canessa, M. (2009). Warehouse design: A structured approach. European
orders to picking stations, the selection and sequence of inventory Journal of Operational Research, 193, 425–436.
pods to satisfy the demands, their parking positions in the storage Banker, S. (2016). Robots in the warehouse: It’s not just Amazon. Forbes. https://
www.forbes.com/sites/stevebanker/2016/01/11/robots- in- the- warehouse- its-
area, and the assignment (along with the path choice) of robots to not- just- amazon/#3aff8e0140b8. (Last access: November 2017).
the specific pod movements all influence each other and can barely Bartholdi III, J. J., & Hackman, S. T. (2016). Warehouse & distribution science: Release
be investigated in an isolated manner. Satisfying the demand for 0.97. Supply Chain and Logistics Institute.
Boysen, N., Boywitz, D., & Weidinger, F. (2018a). Deep-lane storage of time-critical
practical, holistic models will remain a great challenge for future items: One-sided vs. two-sided access. OR spectrum. In press.
research. This aspect is also emphasized by the recent survey paper Boysen, N., Briskorn, D., & Emde, S. (2017). Sequencing of picking orders in mobile
of van Gils et al. (2018) that specifically addresses the combination rack warehouses. European Journal of Operational Research, 259, 293–307.
Boysen, N., Fedtke, S., & Weidinger, F. (2018b). Optimizing automated sorting in
of multiple planning problems.
warehouses: The minimum order spread sequencing problem. European Journal
New objectives: Most of today’s research tackles minimizing the of Operational Research, 270, 386–400.
average processing time per order in a direct or indirect way. How- Boysen, N., & Stephan, K. (2016). A survey on single crane scheduling in auto-
mated storage/retrieval systems. European Journal of Operational Research, 254,
ever, in light of ever-shorter delivery schedules and increasing cus-
691–704.
tomer pretensions, robust warehousing processes are required to Boysen, N., Stephan, K., & Weidinger, F. (2018c). Manual order consolidation with
meet promised delivery dates as best as possible. In this context, put walls: The batched order bin sequencing problem. EURO Journal on Trans-
other objectives than reducing the mean value might be the even portation and Logistic. In press.
Bozer, Y. A., & Kile, J. W. (2008). Order batching in walk-and-pick order picking sys-
better choice. Future research can, for instance, evaluate the impact tems. International Journal of Production Research, 46, 1887–1909.
of minimizing processing time variance or maximum values. Bozer, Y. A., Quiroz, M. A., & Sharp, G. P. (1988). An evaluation of alternative control
Omni-channel retailing is an ongoing business trend that advo- strategies and design issues for automated order accumulation and sortation
systems. Material Flow, 4, 265–282.
cates selling products to customers via multiple different (online Briskorn, D., Emde, S., & Boysen, N. (2017). Scheduling shipments in closed-loop
and offline) distribution channels (Agatz et al., 2008). This consid- sortation conveyors. Journal of Scheduling, 20, 25–42.
erably complicates warehouse operations, because small-sized or- Brynjolfsson, E., Hu, Y. J., & Smith, M. D. (2003). Consumer surplus in the digital
economy: Estimating the value of increased product variety at online book-
ders for customer households have to be jointly assembled with sellers. Management Science, 49, 1580–1596.
large-sized orders for brick-and-mortar stores. Moreover, new con- Bukchin, Y., Khmelnitsky, E., & Yakuel, P. (2012). Optimizing a dynamic order-picking
cepts like click-and-collect, where customers order online and pick process. European Journal of Operational Research, 219, 335–346.
Caputo, A. C., & Pelagagge, P. M. (2006). Management criteria of automated or-
up the products in a store (Hübner et al., 2016), burden the ware-
der picking systems in high-rotation high-volume distribution centers. Industrial
housing processes for brick-and-mortar stores with the same tight Management and Data Systems, 106, 1359–1383.
deadlines as online retailing. The impact of varying order sizes on Çeven, E., & Gue, K. R. (2015). Optimal wave release times for order fulfillment sys-
tems with deadlines. Transportation Science, 51, 52–66.
the warehousing processes has not been appropriately addressed
Chen, C. M., Gong, Y., de Koster, R., & van Nunen, J. A. (2010). A flexible evaluative
in the literature yet. Which warehousing system is best suited to framework for order picking systems. Production and Operations Management,
handle small and large-sized orders together? Or is a mix of sys- 19, 70–82.
tems better suited? Answering these questions is a challenging Chen, T. L., Cheng, C. Y., Chen, Y. Y., & Chan, L. K. (2015). An efficient hybrid algo-
rithm for integrated order batching, sequencing and routing problem. Interna-
task for future research. tional Journal of Production Economics, 159, 158–167.
Return flows: Many online retailers struggle with high return Cimcorp (2011). Cimcorp robot for greater performance in tire manufacture. https://
rates. Rates of 20% and more seem to be rather the norm than www.youtube.com/watch?v=majBgMBCxPM (Last access: November 2017).
Claeys, D., Adan, I., & Boxma, O. (2016). Stochastic bounds for order flow times in
the exception for some product categories (Statista, 2014). In per- parts-to-picker warehouses with remotely located order-picking workstations.
sonal meetings with warehouse managers, we even have heard of European Journal of Operational Research, 254, 895–906.
410 N. Boysen et al. / European Journal of Operational Research 277 (2019) 396–411

Dallari, F., Marchet, G., & Melacini, M. (2009). Design of order picking system. The Herrero-Pérez, D., & Martínez-Barberá, H. (2011). Decentralized traffic control for
International Journal of Advanced Manufacturing Technology, 42, 1–12. non-holonomic flexible automated guided vehicles in industrial environments.
D’Andrea, R., & Wurman, P. (2008). Future challenges of coordinating hundreds Advanced Robotics, 25, 739–763.
of autonomous vehicles in distribution facilities. In Proceedings of the IEEE in- Ho, Y. C., Su, T. S., & Shi, Z. B. (2008). Order-batching methods for an order-pick-
ternational conference technologies for practical robot applications. tePRA 2008 ing warehouse with two cross aisles. Computers & Industrial Engineering, 55,
(pp. 80–83). 321–347.
Daniels, R. L., Rummel, J. L., & Schantz, R. (1998). A model for warehouse order Ho, Y. C., & Tseng, Y. Y. (2006). A study on order-batching methods of order-picking
picking. European Journal of Operational Research, 105, 1–17. in a distribution centre with two cross-aisles. International Journal of Production
Davarzani, H., & Norrman, A. (2015). Toward a relevant agenda for warehousing re- Research, 44, 3391–3417.
search: Literature review and practitioners’ input. Logistics Research, 8, 1–18. Hochrein, S., & Glock, C. H. (2012). Systematic literature reviews in purchasing and
de Koster, R. (2008). Warehouse assessment in a single tour. In M. Lahmar (Ed.), Fa- supply management research: A tertiary study. International Journal of Integrated
cility logistics. approaches and solutions to next generation challenges (pp. 39–60). Supply Management, 7, 215–245.
New York: Taylor & Francis Group. Hong, S., Johnson, A. L., & Peters, B. A. (2012a). Batch picking in narrow-aisle or-
de Koster, R., de Brito, M. P., & van de Vendel, M. A. (2002). Return handling: An der picking systems with consideration for picker blocking. European Journal of
exploratory study with nine retailer warehouses. International Journal of Retail Operational Research, 221, 557–570.
& Distribution Management, 30, 407–421. Hong, S., Johnson, A. L., & Peters, B. A. (2012b). Large-scale order batching in paral-
de Koster, R., Le-Duc, T., & Roodbergen, K. J. (2007). Design and control of ware- lel-aisle picking systems. IIE Transactions, 44, 88–106.
house order picking: A literature review. European Journal of Operational Re- Hong, S., & Kim, Y. (2017). A route-selecting order batching model with the s-shape
search, 182, 481–501. routes in a parallel-aisle order picking system. European Journal of Operational
de Koster, R., Le-Duc, T., & Zaerpour, N. (2012). Determining the number of zones Research, 257, 185–196.
in a pick-and-sort order picking system. International Journal of Production Re- Hsieh, L. F., & Huang, Y. C. (2011). New batch construction heuristics to optimise the
search, 50, 757–771. performance of order picking systems. International Journal of Production Eco-
Enright, J., & Wurman, P. R. (2011). Optimization and coordinated autonomy in mo- nomics, 131, 618–630.
bile fulfillment systems. In Proceedings of the AAAI workshop on automated action Johnson, M. E. (1998). The impact of sorting strategies on automated sortation sys-
planning for autonomous mobile robots (pp. 33–38). tem performance. IIE Transactions, 30, 67–77.
Füßler, D., & Boysen, N. (2017a). Efficient order processing in an inverse order pick- Johnson, M. E., & Meller, R. D. (2002). Performance analysis of split-case sorting
ing system. Computers & Operations Research, 88, 150–160. systems. Manufacturing & Service Operations Management, 4, 258–274.
Füßler, D., & Boysen, N. (2017b). High-performance order processing in picking Jordan, W. C., & Graves, S. C. (1995). Principles on the benefits of manufacturing
workstations. EURO Journal on Transportation and Logistics. In press process flexibility. Management Science, 41, 577–594.
Gademann, A. J. R. M., van den Berg, J. P., & van der Hoff, H. H. (2001). An order Jünemann, R. (1989). Materialfluß und Logistik. Berlin: Springer.
batching algorithm for wave picking in a parallel-aisle warehouse. IIE Transac- Kirks, T., Stenzel, J., Kamagaew, A., & ten Hompel, M. (2012). Zellulare transport-
tions, 33, 385–398. fahrzeuge für flexible und wandelbare intralogistiksysteme. Logistics Journal,
Gagliardi, J. P., Renaud, J., & Ruiz, A. (2012). Models for automated storage and re- 2192(9084), 1–8.
trieval systems: a literature review. International Journal of Production Research, Koch, S., & Wäscher, G. (2016). A grouping genetic algorithm for the order batching
50, 7110–7125. problem in distribution warehouses. Journal of Business Economics, 86, 131–153.
Gallien, J., & Weber, T. (2010). To wave or not to wave? Order release policies for Kulak, O., Sahin, Y., & Taner, M. E. (2012). Joint order batching and picker routing
warehouses with an automated sorter. Manufacturing & Service Operations Man- in single and multiple-cross-aisle warehouses using cluster-based tabu search
agement, 12, 642–662. algorithms. Flexible Services and Manufacturing Journal, 24, 52–80.
Giannikas, V., Lu, W., Robertson, B., & McFarlane, D. (2017). An interventionist strat- Lamballais, T., Roy, D., & de Koster, R. (2017a). Estimating performance in a
egy for warehouse order picking: Evidence from two case studies. International robotic mobile fulfillment system. European Journal of Operational Research, 256,
Journal of Production Economics, 189, 63–76. 976–990.
Gong, Y., & de Koster, R. (2008). A polling-based dynamic order picking system for Lamballais, T., Roy, D., & de Koster, R. (2017b). Inventory allocation in robotic mobile
online retailers. IIE Transactions, 40, 1070–1082. fulfillment systems. Working Paper. Erasmus University Rotterdam.
Gong, Y., & de Koster, R. (2011). A review on stochastic models and analysis of ware- Laudon, K. C., & Traver, C. G. (2007). E-commerce. Pearson/Addison Wesley.
house operations. Logistics Research, 3, 191–205. Le-Duc, T., & de Koster, R. M. (2007). Travel time estimation and order batching in
Grosse, E. H., Glock, C. H., & Ballester-Ripoll, R. (2014). A simulated annealing ap- a 2-block warehouse. European Journal of Operational Research, 176, 374–388.
proach for the joint order batching and order picker routing problem with Li, J., Huang, R., & Dai, J. B. (2017). Joint optimisation of order batching and picker
weight restrictions. International Journal of Operations and Quantitative Manage- routing in the online retailer’s warehouse in china. International Journal of Pro-
ment, 20, 65–83. duction Research, 55, 447–461.
Grosse, E. H., Glock, C. H., Jaber, M. Y., & Neumann, W. P. (2015). Incorporating hu- Litvak, N., & Vlasiou, M. (2010). A survey on performance analysis of warehouse
man factors in order picking planning models: Framework and research oppor- carousel systems. Statistica Neerlandica, 64, 401–447.
tunities. International Journal of Production Research, 53, 695–717. Löffler, M., Boysen, N., Glock, C., & Schneider, M. (2017). Picker routing in AGV-as-
Grosse, E. H., Glock, C. H., & Neumann, W. P. (2017). Human factors in order picking: sisted order picking systems. Working paper DPO-2017-06. RWTH Aachen.
A content analysis of the literature. International Journal of Production Research, Lu, W., McFarlane, D., Giannikas, V., & Zhang, Q. (2016). An algorithm for dynamic
55, 1260–1276. order-picking in warehouse operations. European Journal of Operational Research,
Gu, J., Goetschalckx, M., & McGinnis, L. F. (2007). Research on warehouse opera- 248, 107–122.
tion: A comprehensive review. European Journal of Operational Research, 177, 1– Magazino (2017). Toru: Pick-by-robot for warehouses. https://www.magazino.eu/
21. toru-cube/?lang=en (Last access: November 2017).
Gu, J., Goetschalckx, M., & McGinnis, L. F. (2010). Research on warehouse design Manzini, R., Gamberi, M., & Regattieri, A. (2006). Design and control of an AS/RS.
and performance evaluation: A comprehensive review. European Journal of Op- The International Journal of Advanced Manufacturing Technology, 28, 766–774.
erational Research, 203, 539–549. Marchet, G., Melacini, M., & Perotti, S. (2015). Investigating order picking system
Gue, K. R., Furmans, K., Seibold, Z., & Uludağ, O. (2014). Gridstore: A puzzle-based adoption: a case-study-based approach. International Journal of Logistics Research
storage system with decentralized control. IEEE Transactions on Automation Sci- and Applications, 18, 82–98.
ence and Engineering, 11, 429–438. Matusiak, M., de Koster, R., Kroon, L., & Saarinen, J. (2014). A fast simulated anneal-
Gue, K. R., & Kim, B. S. (2007). Puzzle-based storage systems. Naval Research Logis- ing method for batching precedence-constrained customer orders in a ware-
tics, 54, 556–567. house. European Journal of Operational Research, 236, 968–977.
Gue, K. R., Meller, R. D., & Skufca, J. D. (2006). The effects of pick density on order Matusiak, M., de Koster, R., & Saarinen, J. (2017). Utilizing individual picker skills
picking areas with narrow aisles. IIE Transactions, 38, 859–868. to improve order batching in a warehouse. European Journal of Operational Re-
Guizzo, E. (2008). Three engineers, hundreds of robots, one warehouse – Kiva sys- search, 263, 888–899.
tems wants to revolutionize distribution centers by setting swarms of robots Melacini, M., Perotti, S., & Tumino, A. (2011). Development of a framework for pick-
loose on the inventory. IEEE Spectrum, 45, 26–34. -and-pass order picking system design. The International Journal of Advanced
Hall, R. W. (1993). Distance approximations for routing manual pickers in a ware- Manufacturing Technology, 53, 841–854.
house. IIE Transactions, 25, 76–87. Meller, R. D. (1997). Optimal order-to-lane assignments in an order accumula-
Hübner, A., Holzapfel, A., & Kuhn, H. (2016). Distribution systems in omni-channel tion/sortation system. IIE Transactions, 29, 293–301.
retailing. Business Research, 9, 255–296. Menéndez, B., Bustillo, M., Pardo, E. G., & Duarte, A. (2017a). General variable neigh-
Henn, S. (2015). Order batching and sequencing for the minimization of the total borhood search for the order batching and sequencing problem. European Jour-
tardiness in picker-to-part warehouses. Flexible Services and Manufacturing Jour- nal of Operational Research, 263, 82–93.
nal, 27, 86–114. Menéndez, B., Pardo, E. G., Snchez-Oro, J., & Duarte, A. (2017b). Parallel variable
Henn, S., Koch, S., Doerner, K. F., Strauss, C., & Wäscher, G. (2010). Metaheuristics for neighborhood search for the minmax order batching problem. International
the order batching problem in manual order picking systems. Business Research, Transactions in Operational Research, 24, 635–662.
3, 82–105. Mountz, M. C., D’Andrea, R., LaPlante, J. A., David, P. L. I., Mansfield, P. K., & Ams-
Henn, S., & Schmid, V. (2013). Metaheuristics for order batching and sequencing in bury, B. W. (2008). Inventory system with mobile drive unit and inventory
manual order picking systems. Computers & Industrial Engineering, 66, 338–351. holder. US Patent, 7(402), 018.
Henn, S., & Wäscher, G. (2012). Tabu search heuristics for the order batching prob- Oncan, T. (2015). MILP formulations and an iterated local search algorithm with
lem in manual order picking systems. European Journal of Operational Research, tabu thresholding for the order batching problem. European Journal of Opera-
222, 484–494. tional Research, 243, 142–155.
 N. Boysen et al. / European Journal of Operational Research 277 (2019) 396–411 411

Otto, A., Boysen, N., Scholl, A., & Walter, R. (2017). Ergonomic workplace design in van den Berg, J. P. (1999). A literature survey on planning and control of warehous-
the fast pick area. OR Spectrum, 39, 945–975. ing systems. IIE Transactions, 31, 751–762.
Parikh, P. J., & Meller, R. D. (2008). Selecting between batch and zone order picking van der Gaast, J. P., de Koster, R., & Adan, I. (2018). Analyzing order throughmut
strategies in a distribution center. Transportation Research Part E, 44, 696–719. times in a milkrun picking system. Working Paper, Erasmus University Rotter-
Parikh, P. J., & Meller, R. D. (2010). A note on worker blocking in narrow-aisle or- dam.
der picking systems when pick time is non-deterministic. IIE Transactions, 42, van Gils, T., Ramaekers, K., Caris, A., & de Koster, R. (2018). Designing efficient order
392–404. picking systems by combining planning problems: State-of-the-art classification
Pazour, J. A., & Meller, R. D. (2011). An analytical model for a-frame system design. and review. European Journal of Operational Research, 267, 1–15.
IIE Transactions, 43, 739–752. van Nieuwenhuyse, I., & de Koster, R. B. (2009). Evaluating order throughput time
Petersen, C. G. (20 0 0). An evaluation of order picking policies for mail order com- in 2-block warehouses with time window batching. International Journal of Pro-
panies. Production and Operations Management, 9, 319–335. duction Economics, 121, 654–664.
Qiu, L., Hsu, W. J., Huang, S. Y., & Wang, H. (2002). Scheduling and routing al- Vis, I. F. (2006). Survey of research in the design and control of automated guided
gorithms for AGVs: A survey. International Journal of Production Research, 40, vehicle systems. European Journal of Operational Research, 170, 677–709.
745–760. Žulj, I., Kramer, S., & Schneider, M. (2018). A hybrid of adaptive large neighborhood
Ratliff, H. D., & Rosenthal, A. S. (1983). Order-picking in a rectangular warehouse: search and tabu search for the order-batching problem. European Journal of Op-
A solvable case of the traveling salesman problem. Operations Research, 31, erational Research, 264, 653–664.
507–521. Weidinger, F. (2018). Picker routing in rectangular mixed shelves warehouses. Com-
Roodbergen, K. J., & Vis, I. F. (2006). A model for warehouse layout. IIE Transactions, puters & Operations Research, 95, 139–150.
38, 799–811. Weidinger, F., & Boysen, N. (2018). Scattered storage: How to distribute stock keep-
Roodbergen, K. J., & Vis, I. F. (2009). A survey of literature on automated storage ing units all around a mixed-shelves warehouse. Transportation Science.
and retrieval systems. European Journal of Operational Research, 194, 343–362. Weidinger, F., Boysen, N., & Briskorn, D. (2018). Storage assignment with rack-
Roozbehani, H., & D’Andrea, R. (2011). Adaptive highways on a grid. In Robotics re- moving mobile robots in KIVA warehouses. Transportation Science. In press.
search (pp. (661–680)). Berlin, Heidelberg: Springer. Weidinger, F., Boysen, N., & Schneider, M. (2017). Picker routing in the
Rouwenhorst, B., Reuter, B., Stockrahm, V., van Houtum, G. J., Mantel, R. J., & mixed-shelves warehouses of e-commerce retailers. Working paper.
Zijm, W. H. M. (20 0 0). Warehouse design and control: Framework and litera- Friedrich-Schiller-University Jena.
ture review. European Journal of Operational Research, 122, 515–533. Wruck, S., Vis, I. F., & Boter, J. (2013). Time-restricted batching models and solution
RP Online (2016). Zalando entwickelt pioniergeist. http://www.rp-online.de/nrw/ approaches for integrated forward and return product flow handling in ware-
staedte/moenchengladbach/zalando-entwickelt-pioniergeist-aid-1.6371381 (Last houses. Journal of the Operational Research Society, 64, 1505–1516.
access: November 2017). Wulfraat, M. (2012). Is kiva systems a good fit for your distribution center? an
Russell, M. L., & Meller, R. D. (2003). Cost and throughput modeling of manual and unbiased distribution consultant evaluation. http://www.mwpvl.com/html/kiva_
automated order fulfillment systems. IIE Transactions, 35, 589–603. systems.html (Last access: November 2017).
Scholz, A., Schubert, D., & Wäscher, G. (2017). Order picking with multiple pick- Wurman, P. R., D’Andrea, R., & Mountz, M. (2008). Coordinating hundreds of coop-
ers and due dates simultaneous solution of order batching, batch assignment erative, autonomous vehicles in warehouses. AI Magazine, 29, 9–19.
and sequencing, and picker routing problems. European Journal of Operational Xu, X., Liu, T., Li, K., & Dong, W. (2014). Evaluating order throughput time with
Research, 263, 461–478. variable time window batching. International Journal of Production Research, 52,
Srinivasan, M. M. (1991). Nondeterministic polling systems. Management Science, 37, 2232–2242.
667–681. Yaman, H., Karasan, O. E., & Kara, B. Y. (2012). Release time scheduling and hub
SSI Schäfer (2017). SSI SCHAEFER robo-pickpicking robot. https://www.ssi-schaefer. location for next-day delivery. Operations Research, 60, 906–917.
com/en- de/products/order- picking/automated- order- picking/ssi- schaefer- robo- Yu, J. (2016). Intractability of optimal multirobot path planning on planar graphs.
pick- - - picking- robot- 4612 (Last access: November 2017). IEEE Robotics and Automation Letters, 1, 33–40.
Statista (2014). How high is the average rate of return of goods in your shop Yu, M., & de Koster, R. B. (2009). The impact of order batching and picking area
per year? (Germany, 2014). https://www.statista.com/statistics/327314/share-of- zoning on order picking system performance. European Journal of Operational
returns- in- online- shopping- regarding- fashion- and- consumer- electronics/ (Last Research, 198, 480–490.
access: August 2018). Yuan, R., Cezik, T., & Graves, S. C. (2018). Stowage decisions in multi-zone storage
Statista (2017). Annual retail e-commerce sales growth worldwide from 2014 systems. International Journal of Production Research, 56, 333–343.
to 2020. https://www.statista.com/statistics/288487/forecast- of- global- b2c- e- Yuan, Z., & Gong, Y. Y. (2017). Bot-in-time delivery for robotic mobile fulfillment
commerce-growt/ (Last access: November 2017). systems. IEEE Transactions on Engineering Management, 64, 83–93.
Swisslog (2016). TJ morrishome bargains, united kingdom (Swisslog refer- Zaerpour, N., Yu, Y., & de Koster, R. (2015). Storing fresh produce for fast retrieval
ence). https://www.youtube.com/watch?v=uFO_B5Rj2EQ (Last access: Novem- in an automated compact cross-dock system. Production and Operations Manage-
ber 2017). ment, 24, 1266–1284.
Tappia, E., Roy, D., Melacini, M., & de Koster, R. (2018). Integrated Storage-order pick- Zaerpour, N., Yu, Y., & de Koster, R. (2017). Small is beautiful: A framework for evalu-
ing systems: Technology, performance models, and design insights. Working Pa- ating and optimizing live-cube compact storage systems. Transportation Science,
per. Erasmus University Rotterdam. 51, 34–51.
Techinasia (2017). Alibaba’s robot army. https://www.techinasia.com/video-alibaba- Zhang, J., Wang, X., Chan, F. T., & Ruan, J. (2017). On-line order batching and se-
robot-army (Last access: November 2017). quencing problem with multiple pickers: A hybrid rule-based algorithm. Applied
Tsai, C. Y., Liou, J. J., & Huang, T. M. (2008). Using a multiple-GA method to solve Mathematical Modelling, 45, 271–284.
the batch picking problem: considering travel distance and order due time. In- Zhang, J., Wang, X., & Huang, K. (2016). Integrated on-line scheduling of order
ternational Journal of Production Research, 46, 6533–6555. batching and delivery under b2c e-commerce. Computers & Industrial Engineer-
Valle, C. A., Beasley, J. E., & da Cunha, A. S. (2017). Optimally solving the joint order ing, 94, 280–289.
batching and picker routing problem. European Journal of Operational Research, Zou, B., Gong, Y., Xu, X., & Yuan, Z. (2017). Assignment rules in robotic mobile ful-
262, 817–834. filment systems for online retailers. International Journal of Production Research,
55, 6175–6192.