Roodbergen 2001

This article was downloaded by: [Istanbul Universitesi Kutuphane ve Dok]
On: 21 December 2014, At: 16:21

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954
Registered office: Mortimer House, 37-41 Mortimer Street, London W1T
3JH, UK
International Journal of
Production Research
Publication details, including instructions for
authors and subscription information:
http://www.tandfonline.com/loi/tprs20
Routing methods for

warehouses with multiple
cross aisles
Kees Jan Roodbergen & RenÉde Koster
Published online: 14 Nov 2010.
To cite this article: Kees Jan Roodbergen & RenÉde Koster (2001) Routing methods
for warehouses with multiple cross aisles, International Journal of Production
Research, 39:9, 1865-1883, DOI: 10.1080/00207540110028128
To link to this article: http://dx.doi.org/10.1080/00207540110028128
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the
information (the “Content”) contained in the publications on our platform.
However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness,
or suitability for any purpose of the Content. Any opinions and views
expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the
Content should not be relied upon and should be independently verified with
primary sources of information. Taylor and Francis shall not be liable for any
losses, actions, claims, proceedings, demands, costs, expenses, damages,
and other liabilities whatsoever or howsoever caused arising directly or
indirectly in connection with, in relation to or arising out of the use of the
Content.
This article may be used for research, teaching, and private study purposes.
Any substantial or systematic reproduction, redistribution, reselling, loan,
sub-licensing, systematic supply, or distribution in any form to anyone is
expressly forbidden. Terms & Conditions of access and use can be found at
http://www.tandfonline.com/page/terms-and-conditions
Downloaded by [Istanbul Universitesi Kutuphane ve Dok] at 16:21 21 December 2014
int. j. prod. res., 2001, vol. 39, no. 9, 1865± 1883
Routing methods for warehouses with multiple cross aisles
KEES JAN ROODBERGENy* and RENEÂ DE KOSTERy
This paper considers routing and layout issues for parallel aisle warehouses. In
such warehouses order pickers walk or drive along the aisles to pick products
from storage. They can change aisles at a number of cross aisles. These cross aisles
are usually located at the front and the back of the warehouse, but there can also
be one or more cross aisles at positions in between. We describe a number of

heuristics to determine order picking routes in a warehouse with two or more
cross aisles. To analyse the performance of the heuristics, a branch-and-bound
algorithm that generates shortest order picking routes is used. Performance com-
parisons between heuristics and the branch-and-bound algorithm are given for
various warehouse layouts and order sizes. For the majority of the instances with
more than two cross aisles, a newly developed heuristic appears to perform better
than the existing heuristics. Furthermore, some consequences for layout are dis-
cussed. From the results it appears that the addition of cross aisles to the ware-
house layout can decrease the handling time of the orders by lowering average
travel times. However, adding a large number of cross aisles may increase average
travel times because the space occupied by the cross aisles has to be traversed as
well.
1. Introduction
Warehouses form an important link in the supply chain. Products can be stored
temporarily in warehouses and customer orders can be ® lled by retrieving products
from storage. However, warehousing generally requires a considerable amount of
product handling, which is time consuming. One way to decrease handling time is an
entirely new design of the warehouse. But often it is also possible to decrease hand-
ling time by less radical methods such as changing the operational procedures.
The order picking process is the process of retrieving products from speci® ed
storage locations on the basis of customer orders. The order picking process is in
general one of the most time consuming processes in warehouses and contributes for
a large extent to warehousing costs (see e.g. Tompkins et al. 1996). The productivity
of the order picking process depends on factors such as the storage systems (racks),
the layout and the control mechanisms. Order picking productivity can be improved
by reducing handling time, that is reducing the time needed for picking an order.
This total picking time can be divided roughly into time for driving or walking to
locations (travel time), time for picking the products and time for remaining activ-
ities (such as obtaining a picklist and an empty pick carrier). In warehouses with
Revision received July 2000

{ Erasmus University Rotterdam, Rotterdam School of Management/Faculteit
Bedrijfskunde, PO Box 1738, 3000 DR Rotterdam, The Netherlands.
* To whom correspondence should be addressed. e-mail: k.roodbergen@ fbk.eur.nl
International Journal of Production Research ISSN 0020± 7543 print/ISSN 1366± 588X online # 2001 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
DOI: 10.1080/00207540110028128
1866 K. J. Roodbergen and R. de Koster
manual picking operations, travel time often forms the largest component of the
total picking time (Tompkins et al. 1996).
Several methods can be used to reduce travel times by means of more e cient
control mechanisms. One approach is to determine good order picking routes. The
problem of determining order picking routes consists of ® nding a sequence in which
products have to be retrieved from storage such that the travel distances are as short
as possible. For a warehouse with two cross aisles, one at the front and one at the
back, an e cient algorithm to determine shortest order picking routes has been
developed by RatliŒ and Rosenthal (1983). Heuristics for warehouses with two
cross aisles can be found in Hall (1993). Performance comparisons between optimal
routing and heuristics for this type of warehouse are given in Petersen (1997) and De
Koster and Van der Poort (1998). Another method of reducing travel times is
zoning, that is an order picker picks only that part of an order that is in his or
her assigned zone. Also storage assignment rules can reduce travel times by assigning
products to the correct storage locations. For example, frequently demanded prod-
ucts can be located where they are easily accessible. Important in this respect is the
interaction between the routing method and the storage assignment rule, see for
example, Petersen and Schmenner (1999). Finally, we can think of batching as a
means of reducing travel times. Batching is concerned with combining several (par-
tial) orders in a single order picking route. For batching strategies see Gibson and
Sharp (1992), De Koster et al. (1999) or Ruben and Jacobs (1999).
This paper will focus on routing methods for warehouses with more than two
cross aisles. A preliminary study on heuristic routing in this type of warehouse was
done by Roodbergen and De Koster (1998). They compare three heuristics for a
number of situations, including a narrow-aisle high-bay warehouse where order
picking trucks are used. A routing heuristic, using dynamic programming, for ware-
houses with more than two cross aisles is presented by Vaughan and Petersen (1999).
In this paper we shall extend some heuristics that exist for warehouses with two
cross aisles so that they can be used in warehouses with more than two cross aisles.
Furthermore, we will present a new routing heuristic, called the combined heuristic,
and give some improvements for it, which will be tested under the name combined ‡ .
Like the heuristic of Vaughan and Petersen (1999), the combined heuristic uses
dynamic programming to determine order picking routes. The main diŒerence lies
in the restrictions on the sequencing of the picks. The performance of all heuristics is
compared for the situation of a shelf area where order pickers walk through the
warehouse to pick items using a small pick cart. A branch-and-boun d procedure that
generates shortest order picking routes is used as a benchmark. It is shown that in
most cases, the combined‡ heuristic has a better performance than the other
methods.
Section 2 describes warehouse layout and routing issues and extensions of some
existing routing methods. In § 3 the combined routing heuristic is described to deter-
mine routes in a warehouse with two or more cross aisles. Section 4 compares the
performance of all routing methods. Section 5 comprises the conclusion.
2. Warehouse layout and routing

A graphical sketch of the warehouse layout considered in this paper is given in
® gure 1. The warehouse is rectangular with no unused space and consists of a
number of parallel pick aisles. The warehouse is divided into a number of blocks,
each of which contains a number of subaisles. A subaisle is that part of a pick aisle
Routing for warehouses with multiple cross aisles 1867
Figure 1. Example of a warehouse layout with multiple cross aisles.
that is within one block. The term `aisle’ is used when a statement holds for both pick
aisles and subaisles. At the front and back of the warehouse and between each pair of
blocks, there is a cross aisle. Cross aisles do not contain storage locations, but can be
used to change aisles. Every block has a front cross aisle and a back cross aisle; the
front cross aisle of one block is the back cross aisle of another block, except for the
® rst block. The number of cross aisles equals the number of blocks plus one. This
holds because there is one cross aisle in the front, one in the back and one between
each two adjacent blocks.
Order pickers are assumed to be able to traverse the aisles in both directions and
to be able to change direction within the aisles. The aisles are narrow enough to
allow picking from both sides of the aisle without changing position. See
Goetschalckx and RatliŒ(1988) for issues concerning aisle width. Each order con-
sists of a number of items that are usually spread out over a number of subaisles. We
assume that the items of an order can and will be picked in a single route. Aisle
changes are possible in any of the cross aisles. Picked orders have to be deposited at
the depot, where the picker also receives the instructions for the next route. The
depot is located at the head of the ® rst pick aisle in the front cross aisle. Note that the
location of the depot can potentially in¯ uence the average travel time. Petersen
(1997) evaluated the eŒect of depot location for ® ve picklist sizes and four warehouse
layouts. On average the diŒerence in route length between a depot located in the
corner and a depot in the middle was less than 1% .
In the remainder of this section we describe four diŒerent types of routing. Two
heuristics are based on well known heuristics for a layout with two cross aisles: S-
shape and largest gap. Furthermore, the routing strategy of Vaughan and Petersen
(1999) is described brie¯ y. The fourth routing method consists of ® nding a shortest
route.
2.1. S-shape heuristic

Basically, any subaisle containing at least one pick location is traversed through
the entire length. Subaisles where nothing has to be picked are skipped. In the
following more elaborate description of the heuristic, letters in parentheses corre-
spond to the letters in the example route depicted in ® gure 2(a).
Step 1. Determine the leftmost pick aisle that contains at least one pick location
(called the left pick aisle) and determine the block farthest from the depot
that contains at least one pick location (called the farthest block).
Step 2. The route starts by going from the depot to the front of the left pick aisle (a).
Step 3. Traverse the left pick aisle up to the front cross aisle of the farthest block (b).
Step 4. Go to the right through the front cross aisle of the farthest block until a
subaisle with a pick is reached (c). If this is the only subaisle in this block
with pick locations then pick all items and return to the front cross aisle of
this block. If there are two or more subaisles with picks in this block, then
traverse the subaisle entirely (d).
Step 5. At this point, the order picker is in the back cross aisle of a block, call this
block the current block. There are two possibilities.
Step 5.1. There are picks remaining in the current block (not picked in any
previous step). Determine the distance from the current position to
the leftmost subaisle and the rightmost subaisle of this block with
picks. Go to the closer of these two (e). Entirely traverse this
subaisle (f) and continue with step 6.
Step 5.2. There are no items left in the current block that have to be picked.
In this case, continue in the same pick aisle (i.e. the last pick aisle
that was visited in either step 7 or in this step) to get to the next
cross aisle and continue with step 8.
Step 6. If there are items left in the current block that have to be picked, then
traverse the cross aisle towards the next subaisle with a pick location (g)
and traverse that subaisle entirely (h). Repeat this step until there is exactly
one subaisle left with pick locations in the current block.
Step 7. Go to the last subaisle with pick locations of the current block (i). Retrieve
the items from the last subaisle and go to the front cross aisle of the current
block (j). This step can actually result in two diŒerent ways of travelling
Figure 2. Example routes for four routing methods.
through the subaisle: (i) traversing the subaisle entirely or (ii) entering and
leaving the subaisle from the same side.
Step 8. If the block closest to the depot has not yet been examined, then return to
step 5.
Step 9. Finally, return to the depot (k).
2.2. Largest gap heuristic

The largest gap heuristic basically follows the perimeter of each block, entering
subaisles when needed. The heuristic ® rst goes to the farthest block and then pro-
ceeds block by block to the front of the warehouse. A route resulting from this
heuristic is depicted in ® gure 2(b). Letters in parentheses correspond to the letters
in ® gure 2(b). In this description we say that each subaisle is entered as far as the
`largest gap’ Here we mean a gap is the distance between any two adjacent pick
locations within a subaisle, or between a cross aisle and the nearest pick location.
The largest gap is the largest of all gaps in a subaisle. The largest gap divides the pick
locations in a subaisle into two sets. One set of pick locations is accessed from the
back cross aisle; the other set from the front cross aisle. Note that one or both of the
sets may be empty, making it unnecessary to enter the subaisle from that side.
Step 2. The route starts by going from the depot to the front of the left pick aisle
(a).
Step 3. Traverse the left pick aisle up to the front cross aisle of the farthest block
(b).
Step 4. Go to the right through the front cross aisle of the farthest block until a
subaisle with a pick is reached (c). If this is the only subaisle in this block
with pick locations then pick all items and return to the front cross aisle of
this block. If there are two or more subaisles with picks in this block, then
traverse the subaisle entirely (d).
Step 5. At this point, the order picker is in the back cross aisle of a block, call this
block the current block. There are two possibilities.
Step 5.1. There are picks remaining in the current block (not picked in any
previous step). Determine the subaisle of the current block with
pick locations that is farthest from the current position. Call this
subaisle the last subaisle of the current block. Continue with step
6.
Step 5.2. There are no items left in the current block that have to be
picked. Continue in the same pick aisle (i.e. the last pick aisle
that was visited in either step 7, step 8 or in this step) to get to the
next cross aisle and continue with step 9.
Step 6. Follow the shortest path through the back cross aisle starting at the current
position, visiting all subaisles that have to be entered from the back (e) and
ending at the last subaisle of the current block (f). Each subaisle that is
passed has to be entered up to the largest gap. Note that this step may
require the order picker to walk part of the cross aisle both from left to
right and from right to left (see the example of ® gure 2(b)).
Step 7. Traverse the last subaisle of the current block entirely to get to the front
cross aisle (g).
Step 8. Start at the last subaisle of the current block and move past all subaisles of
the current block that have picks left. Enter these subaisles up to the largest
gap to pick the items (h).
Step 9. If the block closest to the depot has not yet been examined, then return to
step 5.
Step 10. Finally, return to the depot (k).
2.3. Aisle-by-aisle heuristic

This heuristic for warehouses with multiple cross aisles is presented in Vaughan
and Petersen (1999) . Order picking routes resulting from this heuristic visit every
pick aisle exactly once. That is, ® rst all items in pick aisle 1 are picked, then all items
in pick aisle 2, and so on. Dynamic programming is used to determine the best cross
aisles to go from pick aisle to pick aisle.
The order picking route starts at the depot. For every cross aisle i the distance is
calculated that is needed to start at the depot, pick all items in pick aisle 1 and exit
the pick aisle via cross aisle i. If there are m cross aisles, then this results in m
distances each with a corresponding partial order picking route. Now for each
cross aisle j, we determine cross aisle i such that the distance to start at the depot,
pick all items in pick aisle 1, pick all items in pick aisle 2 and exit pick aisle 2 at cross
aisle j; is shortest if we go from pick aisle 1 to pick aisle 2 via cross aisle i. This gives
us again m distances and partial order picking routes. Continuing in a similar
fashion, we determine for each cross aisle j exiting pick aisle 3 the best cross aisle
to go from pick aisle 2 to pick aisle 3. This process is repeated until all pick aisles
have been considered. Then the order picker returns to the depot. An example route
is given in ® gure 2(c).
Note that the algorithm originally assumed that the order picker starts at the
head of the leftmost pick aisle of the warehouse and ends at the rightmost pick aisle
of the warehouse. For reasons of compatibility with the other routing methods, a
minor change in the heuristic was made such that routes start and end at the depot.
2.4. Optimal algorithm

The routing of order pickers in a warehouse is a special case of the Travelling
Salesman Problem. A number of locations have to be visited with the objective to
travel as little as possible. For the Travelling Salesman Problem there is no poly-
nomial-time algorithm known that can ® nd the shortest routes. For warehouses with
two cross aisles, however, an e cient routing algorithm was developed by RatliŒand
Rosenthal (1983). Their method uses dynamic programming to solve the problem.
Extensions of the algorithm to more cross aisles are non-trivial and the number of
equivalence classes and possible transitions increase rapidly. Therefore, we determine
shortest order picking routes in this paper with a branch-and-boun d procedure for
the Travelling Salesman Problem (see e.g. Little et al. (1963)). In ® gure 2(d) an
example route is depicted. The results from the branch-and-boun d procedure will
be used as a benchmark in the performance analysis of the heuristics described in this
paper.
3. Combined heuristic
In practice, it is generally preferred to have a routing method that generates
routes that have a clear and easy to understand structure. Routes having a clear
pattern reduce the time spent by order pickers on searching for locations and reduces
the risk of pick errors. The combined routing method generates such routes. Every
subaisle that contains items is visited exactly once. The route starts and ends at the
depot. The order picker goes through the leftmost pick aisle that contains items
towards the block farthest from the depot that contains items. The subaisles of the
farthest block are visited sequentially from left to right. Then the order picker goes to
the next block (one block closer to the depot). The items in this block are picked.
This process is repeated until all blocks with items have been visited. See ® gure 6(a)
for an example route. The subaisles are either traversed entirely or the order picker
enters and leaves the subaisle from the same side. These choices are made with a
dynamic programming method that will be explained in § 3.2. The construction of a
complete route is discussed in § 3.3.
3.1. DeWnitions
We de® ne the following variables:

k the number of blocks and
n the number of pick aisles.
We denote some physical locations in the warehouse as:
aij the back end of subaisle j in block i, for each block i; i ˆ 1; . . . ; k and for
each subaisle j; j ˆ 1; . . . ; n;
bij the front end of subaisle j in block i, for each block i; i ˆ 1; . . . ; k and for
each subaisle j; j ˆ 1; . . . ; n and
d the depot.
Note that for i ˆ 1; 2; . . . ; k 1 it holds that bij ˆ ai‡1; j . This holds because we
assume that the order pickers walk through the middle of the cross aisles. The
distance from the end of a subaisle to the centre of a cross aisle is administrated
as if it belongs to the subaisle.
For each block we shall use a dynamic programming method. We describe this
dynamic programming method for a single block in § 3.2. In § 3.3 we describe how the
heuristic creates routes, using the dynamic programming method for each single block.
3.2. Dynamic programming method

This section describes a dynamic programming method to route an order picker
through a single block i…i ˆ 1; . . . ; k†. The route starts at the leftmost subaisle that
contains items (`) and ends at the rightmost subaisle that contains items (r). We
de® ne Lj to be a partial route visiting all pick locations in subaisles ` through to j and
we distinguish two classes of partial routes:
Laj , a partial route that ends at the back of subaisle j, and
Lbj , a partial route that ends at the front of subaisle j.
We distinguish two ways to go from subaisle j 1 to subaisle j (see ® gure 3):
ta which goes along the back of the block and
tb which goes along the front of the block.
Furthermore, we distinguish four ways to pick all items in subaisle j. These four
transitions are (see ® gure 3):
t1 , traverse entirely the subaisle,
t2 , do not enter this subaisle at all,
t3 , enter and leave the subaisle from the front of the block, and
t4 , enter and leave the subaisle from the back of the block.
Figure 3. Transitions used by the combined routing heuristic.
Clearly, transition t2 is allowed only if the subaisle does not contain any items.
With Lj ‡ tw we denote that partial route Lj is extended with transition tw
(w ˆ 1; 2; 3; 4; a; b). The function c…:† gives the travel time associated with its argu-
ment, for example c…Lbj ‡ t1 † gives the time needed to walk the partial route Lbj plus
the time needed to walk transition t1 . Note that the transitions contain information
on only how to enter and leave the subaisles. The exact path within the subaisle Ð
and therefore the travel time associated with a transition Ð is dependent on the item
locations within the subaisle under consideration.
Using the potential states, the possible transitions between the states and the
costs (travel time) involved in such transitions, we now give the dynamic program-
ming method. This method will determine a partial route, going through one block.
The construction of the full order picking path, connecting the partial routes for the
individual blocks, will be described in § 3.3.
Step 1
The block under consideration is block i.
If block i is the block farthest from the depot that contains items, then start with
the two partial routes:
La` which starts at node bi` , ends at node ai` and consists of transition t1 and
Lb` which starts and ends at node bi` and consists of transition t3 .
Otherwise, start with two partial routes:
La` which starts and ends at node ai` and consists of transition t4 and
Lb` which starts at node ai` , ends at node bi` and consists of transition t1 .
Step 2
For each consecutive subaisle j …` < j < r† we determine Laj and Lbj as follows.
If subaisle j contains items then:
8
< Laj 1 ‡ ta ‡ t4 if c…Laj 1 ‡ ta ‡ t4 † < c…Lbj 1 ‡ tb ‡ t1 †;
Laj ˆ
: Lb ‡ tb ‡ t1 otherwise;
j 1
(
Lbj 1 ‡ tb ‡ t3 if c…Lbj 1 ‡ tb ‡ t3 † < c…Laj 1 ‡ ta ‡ t1 †;
Lbj ˆ
Laj 1 ‡ ta ‡ t1 otherwise:
If subaisle j does not contain items then:

Laj ˆ Laj 1 ‡ ta ;
Lbj ˆ Lbj 1 ‡ tb :
Step 3
For the last subaisle of the block (subaisle r), we determine
(
b Lbr 1 ‡ tb ‡ t3 if c…Lbr 1 ‡ tb ‡ t3 † < c…Lar 1 ‡ ta ‡ t1 †;
Lr ˆ
Lar 1 ‡ ta ‡ t1 otherwise:
The resulting partial route Lbr will be used to form the complete order picking route.
In this step we no longer need Lar . This is because once all items have been picked in a
block, then we need to go to the front of the block to be able to continue to the next
block (see § 3.3).
3.2.1. Example
Consider one block with three subaisles for which we shall apply the dynamic
programming algorithm. Figure 4 (a) gives the block with pick locations for this
example. This block is assumed to be the block farthest from the depot that contains
items. In total, seven items have to be picked from this block. The length of a
subaisle is 8 m (1 m for each section plus 0.5 m on both sides to go to the centre of
the cross aisle). The distance between two neighbouring subaisles is 4 m. Travel speed
is 1 m s 1 . Figure 4 (b) depicts the situation of ® gure 4 (a) with nodes for the pick
locations and the heads of the subaisles and with edges for the possible travel paths.
Figure 5 visualizes the steps of the dynamic programming algorithm. Note that in
this example ` ˆ 1 and r ˆ 3. All travel times in this example are expressed in
seconds.
Step 1. Since the block under consideration (block i) is assumed to be the block
farthest from the depot, we start with two partial routes La1 and Lb1 . La1 starts
Figure 4. Example situation for the dynamic programming method.

Figure 5. Visualization of the steps taken by the dynamic programming method.
at node bi1 , ends at node ai1 and consists of transition t1 , with associated
travel time c…t1 † ˆ 8 s. Lb1 starts and ends at node bi1 and consists of transi-
tion t3 , with associated travel time c…t3 † ˆ 14 s.
Step 2. We have two possibilities for creating La2 , namely as La1 ‡ ta ‡ t4 (travel time
8 ‡ 4 ‡ 4 ˆ 16 s) or as Lb1 ‡ tb ‡ t1 (travel time 14 ‡ 4 ‡ 8 ˆ 26 s). We
choose the shortest of the two. Thus La2 ˆ La1 ‡ ta ‡ t4 .
Similarly, we have two possibilities for creating Lb2 , namely as Lb1 ‡ tb ‡ t3
(travel time 14 ‡ 4 ‡ 12 ˆ 30 s) or as La1 ‡ ta ‡ t1 (travel time
8 ‡ 4 ‡ 8 ˆ 20 s). We choose the shortest of the two. Thus Lb2 ˆ La1 ‡ ta ‡ t1 .
Step 3. We have two possibilities for creating Lb3 : Clearly, La2 ‡ ta ‡ t1 (travel time
16 ‡ 4 ‡ 8 ˆ 28 s) is faster than Lb2 ‡ tb ‡ t3 (travel time 20 ‡ 4 ‡ 10 ˆ 34 s).
Therefore, Lb3 ˆ La2 ‡ ta ‡ t1 .
This completes the partial route created by the dynamic programming algorithm.
The creation of a full order picking route, going through multiple blocks, will be
discussed in the next section.
3.3. Route construction

So far we have described a dynamic programming method for routing in a single
block. In this section we give a description of the combined routing heuristic, which
uses this dynamic programming method for the individual blocks. The route starts
and ends at the depot. First, the order picker goes to the block farthest from the
depot that contains items via the leftmost pick aisle that contains items. The items in
the leftmost pick aisle are picked and next the items in the farthest block are picked.
Then the order picker moves one block towards the depot and picks all items from
that block. This process is repeated until all blocks with items have been visited.
An example route is depicted in ® gure 6 (a). Each number in this ® gure cor-
responds to one of the numbers of the steps in the description below. Note that
step 6 requires applying the dynamic programming algorithm from § 3.2. It can easily
be seen that we can use the results from the example in § 3.2.1 for the farthest block
in the example of ® gure 6(a). The stepwise procedure for the combined heuristic is
given below.
Step 2. The route starts by going from the depot to the front of the left pick aisle.
Step 3. Traverse the left pick aisle up to the front cross aisle of the farthest block
(block imin ).
Step 4. Set i ˆ imin
Step 5. Determine whether or not block i contains items that have not been picked
in step 3.
Step 5.1. If no items have to be picked in block i: traverse the nearest
subaisle of block i to reach the next block. Continue with step 7.
Step 5.2. If items have to be picked in block i: determine the leftmost sub-
aisle and the rightmost subaisle that contains items (respectively
Figure 6. Route resulting from applying the combined routing method.

subaisle ` and subaisle r), excluding any subaisle that was already
visited in step 3. Go from the current position to the nearest of
these two (jmin ).
Step 6. Apply the dynamic programming method of § 3.2 to block i. If in step 5
jmin ˆ `, then add the partial route resulting from the dynamic programming
algorithm to the order picking path. If jmin ˆ r, then reverse the partial route
resulting from the dynamic programming algorithm (the route then starts in
subaisle r and ends in subaisle `). Add this reversed partial route to the order
picking path. Reversing the path means that the order picker will visit the
subaisles from right to left. The calculations were performed from left to
right.
Step 7. When block k (the block closest to the depot) has been evaluated, the order
picker returns to the depot. Otherwise increase i by 1 and return to step 5.
3.4. Improvements
First of all, consider the routing in the block closest to the depot. The starting
point for routing in this block is determined by the position where the order picker
ends his route through the previous block. This could lead to a route in which aisles
are visited from the left to the right. This implies that the order picker ends his route
somewhere at the right of the front cross aisle. After this, a considerable part of the
front cross aisle has to be traversed before reaching the depot. This can be prevented
by forcing the route to visit aisles from the right to the left in the block closest to the
depot. It can easily be seen that this change in the heuristic will either decrease travel
time or leave the travel time unaltered.
Secondly, consider the path of the order picker from the depot to the farthest
block. This path goes through the leftmost pick aisle with pick locations. However,
one can think of situations where it can be advantageou s to deviate from this path.
In the example of ® gure 6 (a), travel time can be decreased by going through the
second pick aisle towards the back of the warehouse instead of through the ® rst pick
aisle. Stated more generally, we could create routes such that the order picker picks
items from the left x pick aisles on his way to the farthest block and picks items from
the right n x pick aisles when returning to the front. The dynamic programming
method is applied to the left x subaisles of each block on the way to the back and to
the right n x subaisles of each block on the way to the front. By optimizing over x
we obtain a route that is guaranteed to be shorter or at most as long as the route
generated by the original combined heuristic.
We adapt the combined heuristic to incorporate both improvements suggested in
this section and call the result the combined ‡ heuristic. An example route is given in
® gure 6 (b). Note that the two improvements could also be added to the largest gap
and S-shape heuristics. However, the main advantage of these two heuristics is meant
to be their ease of use, which would diminish by such substantial alterations.
Therefore, we do not alter the other heuristics.
4. A comparison of heuristics
This section compares the optimal and heuristic solutions in a practical order
picking system, namely a shelf area. We consider a shelf area where order pickers
walk through the warehouse to pick small items, using a small pick cart. The follow-
ing assumptions are made. The average walking speed in both cross aisles and pick
aisles is 0.6 m s 1 . The centre-to-centre distance between two neighbouring pick aisles
is 2.5 m and no additional time is needed for aisle changing. Cross aisle width is
2.5 m. Picking of items can be performed simultaneously from both sides of a pick
aisle since the aisles are fairly narrow. Order pickers are assumed to walk through the
middle of the pick aisles and cross aisles.
For this type of warehouse we assume the following measures to be representa-
tive. Pick aisle length varies between 10 and 30 m. Each order picker works in a zone
consisting of seven to 15 pick aisles. Each picking route has to visit between 10 and
30 locations. These values are based on observations of numerous actual manual
shelf warehouse operations by the authors. We use the extremes of these values for
our simulation experiments, which give eight diŒerent con® gurations. For each con-
® guration, we generate a number of random orders. The locations of the items in an
order are uniformly and independently distributed over the order picking area. That
is we assume that products are stored randomly in the storage area. No positioning
according to demand frequencies is used. See for example Caron et al. (1998) for
issues involving non-random storage.
For each random order, the route length in a warehouse with two cross aisles is
calculated for the S-shape, largest gap, aisle-by-aisle, combined and combined ‡
heuristics and for the optimal algorithm. Then an additional cross aisle is added
and the route length for each of the routing methods is calculated. Another cross
aisle is added, the route length is calculated, and so on. Averages are taken over the
instances with the same number of cross aisles.
We can distinguish two ways to increase the number of cross aisles. Firstly, we
can ® x the number of storage locations. In this approach the length of the warehouse
will increase if cross aisles are added. Secondly, we could ® x the size of the warehouse
and add cross aisles by losing storage locations. Since usually in design the amount
of storage space is decided in advance, we choose the ® rst approach. This is con-
sistent with Vaughan and Petersen (1999). The minimum number of cross aisles is
two. Such a warehouse with two cross aisles has one cross aisle in the front and one
cross aisle in the back. This restriction is needed since all routing methods assume
that the order picker is able to enter and leave the subaisles both from the front and
from the back. Additional cross aisles are inserted such that the centre distance
between any two adjacent cross aisles is equal. Other ways of cross aisle distribution
may be interesting especially in situations where a storage method other than
random storage is used. The maximum number of cross aisles is 11 for the experi-
ments. Since the shortest pick aisles in the experiments are 10 m, having 11 cross
aisles means that the shortest subaisle encountered in the experiments is 1 m. 1 m can
be considered to be the minimum rack length that is feasible practically in a shelf
warehouse.
For each simulation experiment, the necessary number of replications needs to be
determined such that the estimate for the mean travel time has a relative error
smaller than some ®, for 0 < ® < 1. An approximation for the necessary number
of replications, such that the relative error is smaller than ® with a probability of
1 ¬, is given in Law and Kelton (1991). For all situations considered in this paper,
a replication size of 2000 orders has appeared to be su cient to guarantee a relative
error of at most 1% with a probability of 95% .
Table 1 gives the average travel time for each combination of the eight instances,
ten cross aisle con® gurations and six routing methods. For each con® guration, the
best heuristic is indicated in bold. From the table we can see that the S-shape
Number of cross aisles

No. of No. of
Method aisles Length items 2 3 4 5 6 7 8 9 10 11
Optimal 7 10 10 138.7 129.7 131.5 135.7 141.7 148.0 155.5 162.0 169.6 177.4
7 10 30 186.6 191.4 198.6 207.1 216.3 224.9 235.0 243.2 252.8 261.6
15 10 10 219.6 202.0 201.4 205.2 211.4 218.2 226.7 233.8 242.2 251.0
15 10 30 337.5 314.3 311.6 315.7 324.5 333.4 346.1 355.6 368.7 381.4
7 30 10 269.6 222.9 211.1 209.0 211.4 215.8 221.3 227.4 233.9 240.2
7 30 30 398.3 361.1 342.9 336.5 333.8 334.2 337.0 340.8 345.0 349.7
15 30 10 377.3 308.0 290.9 287.7 289.3 293.5 299.2 305.4 312.0 318.5
15 30 30 665.5 540.6 495.9 479.8 473.3 472.8 475.9 480.7 486.0 491.7
Largest gap 7 10 10 146.6 156.9 164.6 169.3 176.3 181.9 187.4 194.7 201.9 209.3
7 10 30 208.6 240.9 273.9 303.5 330.3 348.6 363.4 379.6 394.4 408.4
15 10 10 227.3 265.2 287.2 296.2 305.7 312.2 317.1 324.6 331.4 338.4
15 10 30 357.5 413.5 484.5 552.1 614.6 660.1 692.3 724.7 750.1 776.3
7 30 10 295.1 259.9 250.7 246.3 247.4 250.3 254.5 259.7 264.9 270.6
7 30 30 451.7 424.7 425.7 435.9 446.9 456.1 464.1 472.2 478.4 488.7
15 30 10 401.0 377.6 379.1 377.2 379.5 382.4 386.6 390.6 395.2 400.6
15 30 30 715.6 646.0 665.4 705.2 746.3 779.9 805.0 826.5 842.8 863.7
S-shape 7 10 10 165.1 145.7 152.6 155.7 161.4 167.7 174.6 181.8 188.6 196.4
7 10 30 203.5 210.3 250.1 253.4 278.3 287.8 301.0 311.9 322.3 332.4
15 10 10 266.2 224.6 245.0 252.6 261.2 270.2 278.9 288.3 295.5 304.6
15 10 30 391.3 359.5 431.1 422.8 478.3 491.1 518.4 539.0 557.8 576.0
7 30 10 353.1 276.2 256.5 245.0 242.5 243.4 247.1 251.7 257.1 262.5
7 30 30 452.0 426.8 438.2 420.1 427.7 423.4 427.0 429.7 432.9 437.1
15 30 10 517.6 376.9 361.0 349.4 347.6 350.1 354.7 360.4 366.4 372.7
15 30 30 833.3 686.0 688.2 636.4 663.4 653.6 666.5 675.2 684.4 695.1
Aisle-by-aisle 7 10 10 148.5 144.3 153.7 164.6 177.5 189.9 205.4 216.4 230.3 245.6
7 10 30 192.1 207.2 227.0 246.7 268.5 289.5 315.7 333.6 357.6 383.8
15 10 10 235.2 220.7 229.4 241.0 255.1 268.9 286.3 298.4 313.6 330.9
15 10 30 356.7 349.3 369.1 392.5 421.3 449.8 486.3 509.8 542.1 578.6
7 30 10 304.7 268.0 268.4 276.5 287.5 299.6 313.0 325.7 339.0 351.8
7 30 30 418.8 413.9 422.8 438.5 457.2 477.7 500.4 522.5 544.1 565.8
15 30 10 427.2 362.0 358.6 366.8 378.3 391.7 406.6 420.6 435.3 449.7
15 30 30 732.7 648.8 642.8 658.6 682.0 709.1 739.7 769.4 798.9 828.9
Combined 7 10 10 148.5 134.6 145.4 151.2 158.2 165.5 172.9 180.4 187.5 195.6
7 10 30 192.1 196.5 236.1 240.4 267.0 277.7 292.2 304.2 315.4 326.3
15 10 10 235.2 208.6 235.3 246.7 257.1 267.2 276.6 286.6 294.1 303.6
15 10 30 356.7 324.6 402.5 399.5 459.0 474.8 504.2 526.8 547.1 566.7
7 30 10 304.7 243.4 235.1 231.2 232.5 236.1 241.5 247.4 253.5 259.7
7 30 30 418.8 386.2 397.4 382.2 394.9 394.7 401.7 407.5 413.8 420.7
15 30 10 427.2 330.1 332.5 331.8 334.8 340.8 347.6 355.0 362.0 369.2
15 30 30 732.7 584.6 605.6 569.7 609.2 608.6 628.2 642.5 656.5 671.3
Combined1 7 10 10 148.5 133.2 136.2 140.2 146.0 152.1 159.4 165.9 173.3 180.8
7 10 30 192.1 196.0 224.7 232.0 244.2 253.2 260.7 268.9 276.6 283.9
15 10 10 235.2 207.0 214.5 220.8 227.6 234.3 242.0 249.3 257.0 264.7
15 10 30 356.7 323.6 373.3 379.6 401.4 415.3 425.5 437.2 447.6 457.2
7 30 10 304.7 235.4 219.3 214.9 216.2 219.8 225.2 230.8 237.2 243.4
7 30 30 418.8 381.7 379.3 365.8 362.7 360.6 361.6 363.6 366.6 370.6
15 30 10 427.2 323.1 305.3 300.5 301.0 304.3 310.1 315.7 322.0 328.5
15 30 30 732.7 577.6 567.6 539.8 538.5 536.7 537.8 541.6 545.2 550.7
Table 1. Average travel time for six routing methods (in seconds).
heuristic never had the best performance of the ® ve heuristics. The largest gap had
the best performance in ® ve situations, each of which has a layout with two cross
aisles. The aisle-by-aisle had the best performance in four situations, of which three
equal the travel time of the combined and combined‡ heuristics. The combined ‡
heuristic gave the best results in 74 of the 80 instances, of which three equal the travel
time of the aisle-by-aisle and combined heuristics. For each of the heuristics the
average calculation time for a single route was less than 0.1 s on a 350 MHz com-
puter.
If we compare the combined and the S-shape heuristic on theoretical grounds, we
can state that for each individual order the combined will give a route that is equal to
or shorter than the S-shape route. This is because S-shape traverses entirely every
subaisle containing items, whereas the combined heuristic chooses between traver-
sing entirely the subaisle or returning to the same side of the subaisle, depending on
which gives the shortest travel time. Thus, the combined heuristic is capable of
generating routes that are exactly the same as those of the S-shape, however, if
possible, it will give shorter order picking routes by returning within subaisles.
The diŒerence in performance between the S-shape and combined depends on the
situation under consideration. For the situations analysed in this paper it holds that
average travel time for the S-shape is at least 7% higher than that for the combined
in warehouses with three cross aisles. The diŒerence between the S-shape and com-
bined tends to zero when increasing the number of cross aisles. The percentage
diŒerence between the combined and S-shape is smaller if the pick density is high
(30 items instead of ten items). This is due to the fact that optimal routes generally
tend to traverse entirely more aisles if the pick density is higher. Therefore S-shape
routes are closer to optimal and the room for improvement is smaller for the com-
bined heuristic.
A second interesting property of the heuristics is the fact that routes of the aisle-
by-aisle, combined and combined‡ heuristics are identical for warehouses with two
cross aisles. This is due to the fact that the heuristics use the same system of dynamic
programming. The aisle-by-aisle creates routes through all blocks and uses one
equivalence class for each cross aisle. The combined and combined ‡ apply dynamic
programming to each block individually and use two equivalence classes, one for
each cross aisle. If the warehouse has two cross aisles (i.e. consists of one block) then
both heuristics use the same two equivalence classes and consequently give the same
routes.
Now let us look at the eŒect of the improvements for the combined heuristic
suggested in § 3.4. The combined and combined ‡ are the same for warehouses with
two cross aisles, because the changes apply only to blocks between the depot and the
farthest block. For situations with three cross aisles, the diŒerence between the
combined and combined ‡ is minor. The improvements are, however, substantial
for situations with more than three cross aisles. The average travel time of the
combined was in these cases up to 24.6% above that of the combined‡ .
In the situations with three or more cross aisles, the combined‡ had the best
performance of the heuristics for all situations except one. For picklists of ten items
and with three or more cross aisles, the diŒerence between the combined‡ and
optimal is less than 7.5% . For the situations considered in this paper, the size of
the gap between the combined‡ and the optimal algorithm varies between 1% and
25% (see table 2). The largest diŒerences occurred for picklists of 30 items. The
situations with seven pick aisles gave a smaller diŒerence between the optimal and
Number of cross aisles

No. of No. of
aisles Length items 2 3 4 5 6 7 8 9 10 11
7 10 10 7.1 2.7 3.6 3.3 3.0 2.8 2.5 2.4 2.2 2.0
7 10 30 2.9 2.4 13.2 12.0 12.9 12.6 10.9 10.6 9.4 8.5
15 10 10 7.1 2.5 6.5 7.6 7.7 7.4 6.8 6.6 6.1 5.5
15 10 30 5.7 2.9 19.8 20.2 23.7 24.6 22.9 22.9 21.4 19.9
7 30 10 13.0 5.6 3.9 2.8 2.2 1.9 1.8 1.5 1.4 1.4
7 30 30 5.2 5.7 10.6 8.7 8.6 7.9 7.3 6.7 6.3 6.0
15 30 10 13.2 4.9 4.9 4.5 4.1 3.7 3.6 3.4 3.2 3.2
15 30 30 10.1 6.8 14.5 12.5 13.8 13.5 13.0 12.7 12.2 12.0
Table 2. Percentage diŒerence in average traveltime between the combined‡ and the optimal
routing method.
combined ‡ than the situations with 15 pick aisles. Generally we can say that the gap
between the optimal and combined‡ tends to be larger if the situation is more
complex, that is more aisles and/or more items.
The observed gap between the best heuristic and the optimal algorithm gives rise
to two diŒerent approaches for further research: (1) develop better heuristics or (2)
use an optimal algorithm for routing order pickers. Both approaches have their
advantages and disadvantages . The heuristics are fairly simple in structure and there-
fore easy to implement in a practical situation. For situations with a large number of
cross aisles, it may be worthwhile to analyse the performance of standard routing
heuristics for the travelling salesman problem. On the other hand, optimal routing
gives signi® cantly shorter routes. However, the logic may be unclear to the order
picker which may cause him to accidentally walk the wrong way or to override the
system and walk the way he thinks is best. Furthermore, in this paper we used a
branch-and-boun d method to calculate the shortest routes. Such a method has
unpredictable computation times, which is an undesirable property for practical
implementations.
From table 1, we can also see the eŒect of adding cross aisles to the layout. In
general we can say that travel time decreases if we change the layout from two to
three cross aisles. Two exceptions occur: (1) with the largest gap heuristic for ware-
houses with short pick aisles and (2) for a small warehouse with many picks. Both
exceptions seem intuitively clear. For the ® rst exception consider that if a warehouse
has short pick aisles then the distance travelled in the cross aisles accounts for a
relatively large amount of the travel time. This is even more the case when using the
largest gap in warehouses with three or more cross aisles, since routes resulting from
this heuristic often traverse cross aisles (except those at the back and front) twice.
Consequently, a large increase in travel time occurs if a third cross aisle is added to
the layout. For the second exception consider a warehouse with a few short pick
aisles and many pick locations. The main advantage of adding cross aisles is that
more possibilities arise to route the order picker. However, if the order picker has to
visit many locations, then entirely traversing pick aisles is close to optimal. Any extra
cross aisle only increases the warehouse size and therefore travel times.
More cross aisles may or may not improve the average time needed to pick an
order. For most situations it holds that adding cross aisles decreases average travel
time up to a certain point after which average travel time starts increasing again. The
optimal number of cross aisles with respect to travel time seems to depend on the
number of pick aisles, the aisle length and the number of items. In particular, aisle
length seems to be important in the sense that longer aisles most often require more
cross aisles. Important for warehouse design is that more cross aisles implies higher
space requirements. Therefore, the cost reductions from adding cross aisles have to
be weighed against increased costs for the building.
5. Concluding remarks
Performances of heuristics in warehouses with two cross aisles have been studied
extensively. In this paper, we have introduced several methods for routing order
pickers in a warehouse with multiple cross aisles. Two methods, the S-shape and
largest gap heuristics, are straightforward extensions of existing methods for ware-
houses with two cross aisles. The aisle-by-aisle heuristic was introduced by Vaughan
and Petersen (1999). The combined and combined‡ heuristics are introduced in this
paper.
For the majority of the situations (74 of 80) evaluated in this paper, the
combined ‡ heuristic had the best performance of the heuristics. The largest gap
was found to be useful in situations with two cross aisles and low pick densities,
which is consistent with Hall (1993).
The performance of the heuristics was also compared to the results of a branch-
and-bound procedure that generates shortest order picking routes. It has to be noted
that the gap between this optimal routing method and the best heuristic varies
substantially. Implementation of the optimal procedure in practical situations
may, however, give rise to problems such as unpredictable computation times. It
could therefore be desirable to improve heuristic performance or ® nd more e cient
methods to calculate shortest routes.
In this paper we considered only situations where products are stored randomly.
Other storage assignment rules may cause a diŒerent ranking among the heuristics.
Furthermore, the positioning of the cross aisles may be an interesting aspect to
consider when using non-random storage, since there will be much activity in the
area with the frequently requested products.
Generally, average travel times decrease when changing the layout from two to
three cross aisles. Two exceptions are the situation of a small warehouse with many
picks for all routing methods and the situations with short pick aisles for the largest
gap. More cross aisles may or may not decrease travel times, depending on the
routing method and the situation under consideration.
References
Caron, F., Marchet, G. and Perego, A., 1998, Routing policies and COI-based storage
policies in picker-to-part systems. International Journal of Production Research, 36, 713±
732.
De Koster, R. and Van der Poort, E., 1998, Routing orderpickers in a warehouse: a
comparison between optimal and heuristic solutions. IIE Transactions, 30, 469± 480.
De Koster, R., Van der Poort, E. and Wolters, M., 1999, E cient orderbatching methods
in warehouses. International Journal of Production Research, 37, 1479± 1504.
Gibson, D. R. and Sharp, G. P., 1992, Order batching procedures. European Journal of
Operational Research, 58, 57± 67.
Goetschalckx, M. and Ratliff, H. D., 1988, Order picking in an aisle. IIE Transactions,
20, 53± 62.
Hall, R. W. H., 1993, Distance approximations for routing manual pickers in a warehouse.
IIE Transactions, 25, 76± 87.
Law, A. M. and Kelton, W. D., 1991, Simulation Modeling and Analysis, 2nd ed. (New
York: McGraw-Hill, Inc.), chapter 9, pp. 522± 581.
Little, J. D. C., Murty, K. G., Sweeney, D. W. and Karel, C., 1963, An algorithm for the
traveling salesman problem. Operations Research, 11, 972± 989.
Petersen, C. G., 1997, An evaluation of order picking routeing policies. International Journal
of Operations and Production Management, 17, 1098± 1111.
Petersen, C. G. and Schmenner, R. W., 1999, An evaluation of routing and volume-based
storage policies in an order picking operation. Decision Sciences, 30, 481± 501.
Ratliff, H. D. and Rosenthal, A. S., 1983, Orderpicking in a rectangular warehouse: A
solvable case of the traveling salesman problem. Operations Research, 31, 507± 521.
Roodbergen, K. J. and De Koster, R., 1998, Routing orderpickers in a warehouse with
multiple cross aisles. In R. J. Graves, L. F. McGinnis, D. J. Medeiros, R. E. Ward and
M. R. Wilhelm (eds.), Progress in Material Handling Research: 1998 (Charlotte, NC:

Material Handling Institute), pp. 451± 467.
Ruben, R. A. and Jacobs, F. R., 1999, Batch construction heuristics and storage assignment
strategies for walk/ride and pick systems. Management Science, 45, 575± 596.
Tompkins, J. A., White, J. A., Bozer, Y. A., Frazelle, E. H., Tanchoco, J. M. A. and
Trevino, J., 1996, Facilities Planning, 2nd edn (New York: John Wiley & Sons, Inc.).
Vaughan, T. S. and Petersen, C. G., 1999, The eŒect of warehouse cross aisles on order
picking e ciency. International Journal of Production Research, 37, 881± 897.

Roodbergen 2001

Uploaded by

Copyright:

Available Formats

Roodbergen 2001

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Roodbergen 2001

Uploaded by

Copyright:

Available Formats

This article was downloaded by: [Istanbul Universitesi Kutuphane ve Dok]

On: 21 December 2014, At: 16:21

Routing methods for

To link to this article: http://dx.doi.org/10.1080/00207540110028128

PLEASE SCROLL DOWN FOR ARTICLE

Routing methods for warehouses with multiple cross aisles

KEES JAN ROODBERGENy* and RENEÂ DE KOSTERy

be one or more cross aisles at positions in between. We describe a number of

Revision received July 2000

2. Warehouse layout and routing

Figure 1. Example of a warehouse layout with multiple cross aisles.

2.1. S-shape heuristic

Figure 2. Example routes for four routing methods.

2.2. Largest gap heuristic

2.3. Aisle-by-aisle heuristic

2.4. Optimal algorithm

We de® ne the following variables:

3.2. Dynamic programming method

Figure 3. Transitions used by the combined routing heuristic.

If subaisle j does not contain items then:

Figure 4. Example situation for the dynamic programming method.

Figure 5. Visualization of the steps taken by the dynamic programming method.

3.3. Route construction

Figure 6. Route resulting from applying the combined routing method.

Number of cross aisles

Number of cross aisles

M. R. Wilhelm (eds.), Progress in Material Handling Research: 1998 (Charlotte, NC:

You might also like