Towards Collaborative Optimisation in a Shared-logistics Environment

for Pickup and Delivery Operations

Timothy Curtois

, Wasakorn Laesanklang

, Dario Landa-Silva

, Mohammad Mesgarpour

and Yi Qu

ASAP Research Group, School of Computer Science, The University of Nottingham, Nottingham NG8 1BB, U.K.

Microlise Ltd, Eastwood, Nottingham, NG16 3AG, U.K.

Keywords:

VRP, PDP, LNS, Horizontal Collaboration, Split-loads.

Abstract:

This paper gives an overview of research work in progress within the COSLE (Collaborative Optimisation in

a Shared Logistics Environment) project between the University of Nottingham and Microlise Ltd. This is an

R&D project that seeks to develop optimisation technology to enable more efﬁcient collaboration in transpor-

tation, particularly real-world operational environments involving pickup and delivery problems. The overall

aim of the project is to integrate various optimisation techniques into a framework that facilitates collaboration

in a shared freight transport logistics environment with the overall goal of reducing empty mileage.

1 INTRODUCTION

This paper provides an overview of the research work

being undertaken as part of the COSLE (Collabora-

tive Optimisation in a Shared Logistics Environment)

project. This is an R&D project between the Uni-

versity of Nottingham and Microlise Ltd in the UK.

The overall objective of the project is to develop op-

timisation technology to enable more efﬁcient colla-

boration in transportation. According to a recent re-

port from the Institution of Mechanical Engineers in

2016 (Oldham, 2016), up to 30% of all commercial

vehicles on UK roads travel empty, which leads to

around 150m wasted road miles, 200,000 additional

truck journeys, increased road congestion and about

200,000 tonnes of unnecessary CO

emissions. One

way to improve this situation is by facilitating col-

laboration between carriers. The improved coopera-

tion will reduce the total distances that vehicles travel

without loads (so called empty miles), increase vehi-

cle utilisation metrics and decrease distribution costs.

Already there have been several successful applicati-

ons of increased cooperation in transportation. (Crui-

jssen et al., 2007a) modelled a joint route planning

problem, used a benchmark case and reported that

30.7% of total distribution costs are saved. (Ergun

et al., 2007) proposed a lane covering problem and

solved it using heuristics. Results showed that the

saving range from about 5.5% to a little over 13%,

again using real data. (Frisk et al., 2010) presen-

ted a case study of horizontal collaboration in tactical

transportation planning between eight forest compa-

nies and the results showed up to 14.2% of the trans-

portation cost saved. (P

erez-Bernabeu et al., 2015)

discussed horizontal collaboration in road transporta-

tion and presented numerical analysis based on a set

of well-known benchmarks for the Multidepot Vehi-

cle Routing Problem. The average cost reduction ran-

ges from 5% to 90% depending on the geographical

distribution of customers with respect to their trans-

port service providers. It has been proved by rese-

archers that saving of costs and increase of resource

utilization can be attained throughout horizontal col-

laboration (Cruijssen et al., 2007b; Wang and Kopfer,

2014).

The goal of the COSLE project is to develop

an innovative service to enable collaboration in a

shared freight transport logistics environment to re-

duce empty freight runs. As part of this, three sub-

projects related to scheduling and optimisation have

been identiﬁed and are currently being undertaken by

project team. This position paper provides an over-

view of these sub-projects and the progress achieved

so far. The ﬁrst sub-project is to develop a metho-

dology to tackle pickup and delivery problems with

time windows and other real-world constraints. A

metaheuristic approach has been developed and tes-

ted on a benchmark data set. This work and a sum-

mary of the results is described further in Section 2.

The second sub-project is a method to enable opti-

Curtois T., Laesanklang W., Landa-Silva D., Mesgarpour M. and Qu Y.

Towards Collaborative Optimisation in a Shared-logistics Environment for Pickup and Delivery Operations.

DOI: 10.5220/0006291004770482

In Proceedings of the 6th International Conference on Operations Research and Enterprise Systems (ICORES 2017), pages 477-482

ISBN: 978-989-758-218-9

477

mal load-splitting within routes and schedules. This

is presented in Section 3 as well as a review of related

research. The third sub-project, discussed in Section

4, is to develop a methodology for assigning new cu-

stomers within existing routes. The outcomes from

these three sub-projects will be integrated into a fra-

mework that will aim to enable more efﬁcient colla-

boration in transportation under real-world operatio-

nal conditions. The overall approach is to develop

an optimisation engine for tackling pickup and deli-

very routing scenarios in which various transportation

operators are willing to collaborate in order to incre-

ase the overall utilisation of vehicles by reducing the

number of empty runs.

2 A HYBRID METAHEURISTIC

FOR PDP

The pickup and delivery problem (PDP) is a widely

occurring vehicle routing problem. Similar to other

vehicle routing problems it often contains window

and capacity constraints. Unlike the general vehicle

routing problem however PDP also includes pairing

and precedence constraints. The pairing constraint

is to ensure that a pickup customer and its associa-

ted delivery customer are both serviced by the same

vehicle. The precedence constraint does not allow

a delivery customer to be visited before its associa-

ted pickup customer. The techniques being investi-

gated in this project are for tackling pickup and deli-

very problems which also contain window and capa-

city constraints as well as other real world constraints

such as driver working time regulations and break re-

quirements. The objective function, similar to other

problems, requires the minimisation of the number

of vehicles used and the total distance travelled. Ot-

her optional objectives allow the minimisation of total

driver hours, and a proﬁt maximisation objective for

problems in which some customers may be optionally

serviced and have an associated completion cost.

Various heuristic and exact methods have been

proposed for PDP. Each method has advantages and

disadvantages. The exact methods, although extre-

mely effective on smaller instances, appear to still be

difﬁcult to apply to the largest instances. Metaheu-

ristics however have been shown to scale much more

easily to larger instances although are easily beaten on

smaller instances. They can also provide no informa-

tion on solution optimality or even bounds. However

a recent survey (Hall and Partyka, 2016) suggests that

most industrial vehicle routing packages are still hea-

vily biased towards using metaheuristics.

Of the exact methods published, many are versi-

ons of the column generation and branch and price

framework (Dumas et al., 1991; Ropke and Cor-

deau, 2009; Savelsbergh and Sol, 1998; Venkates-

han and Mathur, 2011; Xu et al., 2003) or less com-

monly, branch and cut (Lu and Dessouky, 2004; Ru-

land and Rodin, 1997). Examples of metaheuristics

include (Bent and Van Hentenryck, 2006; Li and Lim,

2003; Nagata and Kobayashi, 2010; Nanry and Bar-

nes, 2000; Ropke and Pisinger, 2006). Metaheuris-

tics have also been applied to less common variants of

PDP (Cherkesly et al., 2015; Kammarti et al., 2004;

Masson et al., 2013). Several survey papers are also

available (Berbeglia et al., 2007; Parragh et al., 2008;

Savelsbergh and Sol, 1995).

The hybrid method that has been developed here

combines Local Search, Large Neighbourhood Se-

arch (LNS) and Guided Ejection Search (GES). It

works in several phases. In the ﬁrst phase, local se-

arch using four different neighbourhood operators is

used to create an initial solution. The operators used

are:

1. Inserting unassigned customers into routes.

2. Moving a customer from one route to another.

3. Swapping customers between routes.

4. Moving a customer from one route to a second

route and simultaneously moving a customer from

a second route to a third route.

Even on the largest instances the local search

phase is very fast but the solutions produced are ne-

arly always quite sub-optimal. The next phase uses

Guided Ejection Search in an attempt to minimise

the total number of routes in the solution. The GES

implementation is based on (Nagata and Kobayashi,

2010). It works by iteratively, randomly selecting

a route, un-assigning all customers in the route and

then attempting to re-insert these customers in exis-

ting routes. If it is unable to insert a customer it ejects

one or more customers from a route to enable it to in-

sert the customer. The ejected customer(s) are then

added to the list of customers still to be inserted. Af-

ter the ejection the solution is perturbed by randomly

moving or swapping already assigned customers bet-

ween routes. This helps to insert un-assigned custo-

mers and also prevents inﬁnite loops. This is repeated

for a number of iterations or until all customers have

been inserted. If all customers are inserted then the

process is repeated to try and remove another route.

Otherwise the original solution is restored.

After the GES phase, a Large Neighbourhood Se-

arch is applied to try and improve the other objecti-

ves (total distance for the benchmark instances). The

LNS is a simpliﬁed version of the adaptive LNS of

(Ropke and Pisinger, 2006). One of the simpliﬁcati-

ICORES 2017 - 6th International Conference on Operations Research and Enterprise Systems

478

Table 1: Summary of Results.

Customers Instances New best

knowns

Equal best

knowns

50 56 0 56

100 60 7 35

200 60 22 19

300 60 33 6

400 60 45 5

500 58 35 4

ons was to remove the adaption procedure which was

shown by the authors to have only a small percen-

tage beneﬁt. Our results also conﬁrmed that excellent

solutions could still be obtained without the adaption

procedure. Another modiﬁcation was to replace a si-

mulated annealing heuristic with a late acceptance hill

climbing heuristic (Burke and Bykov, 2012). The mo-

tivation for this was to remove the number of parame-

ters that required setting. The algorithm operates by

iteratively un-assigning a small number of customers

and then heuristically attempting to re-insert them but

in a lower cost conﬁguration. If it is unable to re-

insert them in a better way then it restores the original

solution and selects a new set of customers for remo-

val and re-insertion. This process is iteratively repea-

ted. The customers for removal are selected randomly

or via the Shaw heuristic (Shaw, 1998) which selects

customers that are similar in terms of location, time

windows and order size. The insertion heuristics are

based on the regret assignment heuristic (Ropke and

Pisinger, 2006).

After the completion of the LNS phase, if there is

time remaining then the best known solution is pertur-

bed by randomly moving or swapping customers bet-

ween routes. The three phases are then applied again

to the perturbed solution. This whole process is repe-

ated until a ﬁxed time limit is reached. At which point

the best known solution is returned.

In order to evaluate the efﬁcacy of the algorithm

it was applied to the well-known benchmark problem

instances of (Li and Lim, 2003)

. These instances are

divided by size into six groups, ranging from 50 cu-

stomers up to 500 customers. The results are sum-

marized in Table 1. In Table 1. the column “New

best solution” indicates the number of instances that

our algorithm was able to ﬁnd a new best known so-

lution. The column “equal best knowns” indicates the

number of instances on which the algorithm was able

to equal the current best known solution for that in-

stance.

Available at http://www.sintef.no/projectweb/top/pdptw

/li-lim-benchmark/

3 LOAD SPLITTING

Often job loads can be partially collected and deli-

vered multiple times provided they are completed in

entirety within the given time windows. This option

allows the possibility of split loads to be used opera-

tionally.

The obvious case is when a requested demand ex-

ceeds vehicle capacity. Demands in this case must

be split before the optimisation process. The research

question in this case is how to split the requested de-

mands so that the split loads aid the optimisation pro-

cess. Some of the papers in the literature saw this case

as a part of pre-processing in optimisation problem.

The other case is to gain additional savings in the

operational plan. The literature shows split loads can

reduce operational costs (Andersson et al., 2011; No-

wak et al., 2009). The beneﬁts of split loads were

subject to problem characteristics such as load size,

stopping cost, and frequency of loads having common

pickup and delivery locations.

The transportation problem with split loads arose

in the generic vehicle routing problem (VRP) in (Dror

and Trudeau, 1989). The idea was to relax the VRP

so that a customer can be visited more than once. A

k-split interchange is also proposed as a heuristic pro-

cedure to split a demand into multiple loads. The so-

lution after the split procedure was expected to have

cost reduction from the generic problem.

(Dror and Trudeau, 1989) used a heuristic process

to tackle a split load problem in two stages: construct

a solution to the generic VRP; and apply k-split in-

terchange and improvement routine to get a solution

to the split load problem. The k-split interchange was

also used as a move in a tabu search algorithm for

the split delivery vehicle routing problem by (Archetti

et al., 2006). They describe k-split interchange in two

main procedures:

1. Remove a demand i from all routes where it is

visited; and

2. Find a route subset R where the summation of re-

maining capacity is larger than the demand i so

that the demand i is split into every route in the

route subset R.

The route subset R should have the least insertion

cost.

A randomised granular tabu search heuristic was

used to solve the split delivery vehicle routing pro-

blem by (Berbotto et al., 2014). The method builds a

granular neighbourhood to reduce the computational

time required to explore solution neighbourhood.

Pickup and delivery with split loads was tackled

by a heuristic where two route segments of different

routes can visit the same pickup and delivery demand

Towards Collaborative Optimisation in a Shared-logistics Environment for Pickup and Delivery Operations

479

by (Nowak et al., 2008). The demand can be carried

by two vehicles.

A requirement of split delivery in simultaneous

pickup and delivery arose in automobile industries in

(Tang et al., 2009). At the supplier location, a truck

must deliver empty bins to the pickup locations in or-

der to pick up the full bins. In the same way, at the

manufacturer, the truck must deliver full bins and pick

up empty bins. Bins are cycled between the manufac-

turer and the suppliers.

Coordination split delivery can also beneﬁt re-

tailers in maintaining stock levels (Li et al., 2011).

This approach can reduce retailer inventory costs

while the transportation cost remains the same. A

similar application can apply to the natural disaster

relief distribution problem (Wang et al., 2014). The

goal of this case was to distribute sufﬁcient aid to the

disaster areas. A disaster area can be visited multiple

times. A full review on split delivery transportation

problems can also be found in (Archetti and Speranza,

2012).

The closest application to the split pickup and de-

livery in the collaborative logistics environment con-

sidered here is the pickup and delivery problem with

split load proposed by (Nowak et al., 2008). There-

fore, we adopt their split load creation procedure and

apply it to the large neighbourhood search method.

The same procedure can also be used to split demand

that exceeds vehicle capacity. The split load creation

procedure works similarly to the k-split interchange

procedure. The procedure is as follows:

1. Find a segment i to split;

2. Find a segment j the load should move to;

3. Split the load in the segment i where the ﬁrst split

load is equal to the excess capacity of segment j,

and the second split load is the remainder;

4. Move the split load to segment j and the remain-

der load is kept in the segment i;

5. Perform the search heuristic.

The proposed approach in this sub-project is to

implement move operators within LNS in order to

handle split loads. This includes the delete split ope-

rator, exchange split operator, etc. These operators

were applied to VRP (Berbotto et al., 2014). The de-

lete split operator removes a set of split loads to be-

come a full demand load.

The exchange split operator swaps the load split-

ting position in a selected route. In VRP, the idea was

to swap the position to split a load while maintaining

vehicle capacity. Suppose we have a load i and a load

j where load i is split into two smaller loads and load j

and one of the split loads of i are assigned to a vehicle.

This operator relocates the split position from load i

to load j which results in the vehicle taking the full

load i and a partial load j. For our PDP, the operator

starts from selecting an interval where a vehicle has

split loads in their ﬁll. The exchange split operator

will:

1. Delete the split of a demand; and

2. Apply a split to one of the other demands.

The demands that the exchange split operator can se-

lect must be the ﬁll in the selected interval only. The

operator keeps the visit order of the selected route but

may change the order of the routes that operate on the

split demands.

This splitting heuristic and move operators will be

integrated into the hybrid metaheuristic for PDP out-

lined in Section 2 and applied to large real-world in-

stances. The beneﬁts of providing splitting options

will then be analysed. Optionally, the splitting heu-

ristic can be adapted to increase vehicle empty space

available for taking advantage of new collaborative

opportunities. In the same way, the heuristic can split

the collaborative jobs so that they can be inserted into

the existing routes.

4 CUSTOMER INSERTION INTO

EXISTING ROUTES

Another requirement in collaborative transport opera-

tions is to be able to insert new customers into exis-

ting routing plans. It might be that the ordering of the

customers in the routes of the existing solution can-

not be changed but their arrival times could be adjus-

ted provided that their window constraints are still re-

spected. Existing customers must also remain within

their current routes. Hence, the hybrid LNS+GES al-

gorithm outlined in Section 2 cannot simply be app-

lied to a new instance which includes the new custo-

mers. Instead, a separate mechanism is being develo-

ped to insert the new customers.

To the best of our knowledge this problem has

little or no previously published research articles.

Modiﬁed but similar versions of the problem do so-

metimes appear as sub-problems in methodologies for

vehicle routing problems though. For example, re-

lated problems are solved using branch and bound

and constraint programming algorithms in (Bent and

Van Hentenryck, 2006; Shaw, 1998). In (Ropke and

Pisinger, 2006) also use a heuristic method to solve

a version of the sub-problem for pickup and delivery

with time window problems.

For the insertion problem considered here two se-

parate objectives for two different scenarios are pro-

posed:

ICORES 2017 - 6th International Conference on Operations Research and Enterprise Systems

480

1. Maximise the number of customers inserted.

2. Maximise proﬁt. In this scenario customers are

assigned values (revenue) and a cost is calculated

based on total solution distance and/or total driver

hours.

The insertion problem can be formulated as an in-

teger programming problem and solved using a mat-

hematical programming solver. We will also be in-

vestigating and comparing heuristic methods and al-

ternative exact methods to establish the computation

time/efﬁciency trade-off for the different approaches.

For example, the hybrid LNS+GES algorithm already

contains an existing insertion algorithm in the form

of the regret heuristic. Greedy insertion heuristics

are also feasible options. Other insertion algorithms

that we will be developing and testing are the branch

and bound approaches in (Shaw, 1998) and (Bent and

Van Hentenryck, 2006).

To analyse the algorithms a testing framework has

been created to allow us to efﬁciently repeat and re-

produce the results. The test instances were created

by taking existing instances, removing sets of cus-

tomers and solving the reduced instances using the

LNS+GES algorithm. The original instance is then

used with this initial solution to form a new insertion

instance. This procedure is repeated with different pa-

rameter settings to generate a large set of test instan-

ces to apply the algorithms to.

Another motivation for developing and analysing

several insertion methods is to investigate whether a

more efﬁcient and effective method can be developed

for the LNS algorithm. If so then it is possible that

the LNS algorithm can be further improved by incor-

porating the new insertion algorithm.

ACKNOWLEDGEMENTS

We thank Innovate UK for funding the COSLE pro-

ject (grant 102037).

REFERENCES

Andersson, H., Christiansen, M., and Fagerholt, K. (2011).

The maritime pickup and delivery problem with time

windows and split loads. INFOR: Information Sys-

tems and Operational Research, 49(2):79–91.

Archetti, C. and Speranza, M. G. (2012). Vehicle routing

problems with split deliveries. International transacti-

ons in operational research, 19(1-2):3–22.

Archetti, C., Speranza, M. G., and Hertz, A. (2006). A tabu

search algorithm for the split delivery vehicle routing

problem. Transportation Science, 40(1):64–73.

Bent, R. and Van Hentenryck, P. (2006). A two-stage hybrid

algorithm for pickup and delivery vehicle routing pro-

blems with time windows. Computers & Operations

Research, 33(4):875–893.

Berbeglia, G., Cordeau, J.-F., Gribkovskaia, I., and Laporte,

G. (2007). Static pickup and delivery problems: a

classiﬁcation scheme and survey. TOP, 15(1):1–31.

Berbotto, L., Garc

ıa, S., and Nogales, F. J. (2014). A rand-

omized granular tabu search heuristic for the split de-

livery vehicle routing problem. Annals of Operations

Research, 222(1):153–173.

Burke, E. K. and Bykov, Y. (2012). The late acceptance

hill-climbing heuristic. University of Stirling, Tech.

Rep.

Cherkesly, M., Desaulniers, G., and Laporte, G. (2015). A

population-based metaheuristic for the pickup and de-

livery problem with time windows and lifo loading.

Computers & Operations Research, 62:23–35.

Cruijssen, F., Br

aysy, O., Dullaert, W., Fleuren, H., and

Salomon, M. (2007a). Joint route planning under

varying market conditions. International Journal

of Physical Distribution & Logistics Management,

37(4):287–304.

Cruijssen, F., Dullaert, W., and Fleuren, H. (2007b). Hori-

zontal cooperation in transport and logistics: a litera-

ture review. Transportation journal, pages 22–39.

Dror, M. and Trudeau, P. (1989). Savings by split delivery

routing. Transportation Science, 23(2):141–145.

Dumas, Y., Desrosiers, J., and Soumis, F. (1991). The

pickup and delivery problem with time windows. Eu-

ropean Journal of Operational Research, 54(1):7–22.

Ergun, O., Kuyzu, G., and Savelsbergh, M. (2007). Redu-

cing truckload transportation costs through collabora-

tion. Transportation Science, 41(2):206–221.

Frisk, M., G

othe-Lundgren, M., J

ornsten, K., and

onnqvist, M. (2010). Cost allocation in collaborative

forest transportation. European Journal of Operatio-

nal Research, 205(2):448–458.

Hall, R. and Partyka, J. (2016). Vehicle routing soft-

ware survey: Higher expectations drive transforma-

tion. ORMS-Today, 43(1).

Kammarti, R., Hammadi, S., Borne, P., and Ksouri, M.

(2004). A new hybrid evolutionary approach for the

pickup and delivery problem with time windows. In

Systems, Man and Cybernetics, 2004 IEEE Interna-

tional Conference on, volume 2, pages 1498–1503.

IEEE.

Li, H. and Lim, A. (2003). A metaheuristic for the

pickup and delivery problem with time windows. In-

ternational Journal on Artiﬁcial Intelligence Tools,

12(02):173–186.

Li, J., Chu, F., and Chen, H. (2011). Coordination of

split deliveries in one-warehouse multi-retailer distri-

bution systems. Computers & Industrial Engineering,

60(2):291–301.

Lu, Q. and Dessouky, M. (2004). An exact algorithm for the

multiple vehicle pickup and delivery problem. Trans-

portation Science, 38(4):503–514.

Masson, R., Lehu

e, F., and P

eton, O. (2013). An adap-

tive large neighborhood search for the pickup and de-

Towards Collaborative Optimisation in a Shared-logistics Environment for Pickup and Delivery Operations

481

livery problem with transfers. Transportation Science,

47(3):344–355.

Nagata, Y. and Kobayashi, S. (2010). Guided ejection se-

arch for the pickup and delivery problem with time

windows. In European Conference on Evolutionary

Computation in Combinatorial Optimization, pages

202–213. Springer.

Nanry, W. P. and Barnes, J. W. (2000). Solving the pickup

and delivery problem with time windows using re-

active tabu search. Transportation Research Part B:

Methodological, 34(2):107–121.

Nowak, M., Ergun, O., and White, C. C. (2009). An empi-

rical study on the beneﬁt of split loads with the pickup

and delivery problem. European Journal of Operatio-

nal Research, 198(3):734–740.

Nowak, M., Ergun,

O., and White III, C. C. (2008). Pickup

and delivery with split loads. Transportation Science,

42(1):32–43.

Oldham, P. (2016). UK Freight: in for the long haul. Insti-

tution of Mechanical Engineers.

Parragh, S. N., Doerner, K. F., and Hartl, R. F. (2008). A

survey on pickup and delivery problems. Journal f

Betriebswirtschaft, 58(1):21–51.

erez-Bernabeu, E., Juan, A. A., Faulin, J., and Barrios,

B. B. (2015). Horizontal cooperation in road trans-

portation: a case illustrating savings in distances and

greenhouse gas emissions. International Transactions

in Operational Research, 22(3):585–606.

Ropke, S. and Cordeau, J.-F. (2009). Branch and cut and

price for the pickup and delivery problem with time

windows. Transportation Science, 43(3):267–286.

Ropke, S. and Pisinger, D. (2006). An adaptive large neig-

hborhood search heuristic for the pickup and delivery

problem with time windows. Transportation science,

40(4):455–472.

Ruland, K. and Rodin, E. (1997). The pickup and delivery

problem: Faces and branch-and-cut algorithm. Com-

puters & mathematics with applications, 33(12):1–13.

Savelsbergh, M. and Sol, M. (1998). Drive: Dynamic rou-

ting of independent vehicles. Operations Research,

46(4):474–490.

Savelsbergh, M. W. and Sol, M. (1995). The general

pickup and delivery problem. Transportation science,

29(1):17–29.

Shaw, P. (1998). Using constraint programming and local

search methods to solve vehicle routing problems. In

International Conference on Principles and Practice

of Constraint Programming, pages 417–431. Springer.

Tang, G., Ning, A., Wang, K., and Qi, X. (2009). A practi-

cal split vehicle routing problem with simultaneous

pickup and delivery. In Industrial Engineering and

Engineering Management, 2009. IE&EM’09. 16th In-

ternational Conference on, pages 26–30. IEEE.

Venkateshan, P. and Mathur, K. (2011). An efﬁ-

cient column-generation-based algorithm for solving

a pickup-and-delivery problem. Computers & Opera-

tions Research, 38(12):1647–1655.

Wang, H., Du, L., and Ma, S. (2014). Multi-objective open

location-routing model with split delivery for optimi-

zed relief distribution in post-earthquake. Transpor-

tation Research Part E: Logistics and Transportation

Review, 69:160–179.

Wang, X. and Kopfer, H. (2014). Collaborative transpor-

tation planning of less-than-truckload freight. OR

spectrum, 36(2):357–380.

Xu, H., Chen, Z.-L., Rajagopal, S., and Arunapuram, S.

(2003). Solving a practical pickup and delivery pro-

blem. Transportation science, 37(3):347–364.

ICORES 2017 - 6th International Conference on Operations Research and Enterprise Systems

482