Ya Regal
Ya Regal
Invited Review
a r t i c l e i n f o a b s t r a c t
Article history: A number of technological advances have led to a renewed interest in dynamic vehicle routing problems.
Received 27 September 2011 This survey classifies routing problems from the perspective of information quality and evolution. After
Accepted 21 August 2012 presenting a general description of dynamic routing, we introduce the notion of degree of dynamism, and
Available online 30 August 2012
present a comprehensive review of applications and solution methods for dynamic vehicle routing
problems.
Keywords: Ó 2012 Elsevier B.V. All rights reserved.
Transportation
Combinatorial optimization
Dynamic vehicle routing
0377-2217/$ - see front matter Ó 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.ejor.2012.08.015
2 V. Pillac et al. / European Journal of Operational Research 225 (2013) 1–11
Table 1
Taxonomy of vehicle routing problems by information evolution and quality.
Information quality
Deterministic input Stochastic input
Information evolution Input known beforehand Static and deterministic Static and stochastic
Input changes over time Dynamic and deterministic Dynamic and stochastic
planning a trip back to the depot or skipping a customer. Applica- able to track and manage their fleet in real time and cost effec-
tions in this category do not require any technological support. tively. While traditionally a two-step process (i.e., plan-execute),
Uncertainty may affect any of the input data, yet the three most vehicle routing can now be done dynamically, introducing greater
studied cases are [33]: stochastic customers, where a customer opportunities to reduce operational costs, improve customer ser-
needs to be serviced with a given probability [15,147]; stochastic vice, and reduce environmental impact.
times, in which either service or travel times are modeled by ran- The most common source of dynamism in vehicle routing is the
dom variables [79,85,146]; and lastly, stochastic demands online arrival of customer requests during the operation. More spe-
[31,37,86,97,98,126,128]. Further details on the static stochastic cifically, requests can be a demand for goods [2,62,70,71,75,
vehicle routing can be found in the reviews by Bertsimas and Sim- 99,101,144] or services [7,11,17,52,89,141]. Travel time, a dynamic
chi-Levi [33], Cordeau et al. [16], and Gendreau et al. [53]. component of most real-world applications, has been recently ta-
In dynamic and deterministic problems, part or all of the input is ken into account [1,6,28,48,64,65,94,108,136,139,153]; while ser-
unknown and revealed dynamically during the design or execution vice time has not been explicitly studied (but can be added to
of the routes. For these problems, vehicle routes are redefined in an travel time). Finally, some recent work considers dynamically re-
ongoing fashion, requiring technological support for real-time vealed demands for a set of known customers [104,105,126,128]
communication between the vehicles and the decision maker and vehicle availability [92,93,103], in which case the source of
(e.g., mobile phones and global positioning systems). This class of dynamism is the possible breakdown of vehicles. In the following
problems are also referred to as online or real time by some authors we use the prefix ‘‘D-’’ to label problems in which new requests ap-
[77]. pear dynamically.
Similarly, dynamic and stochastic problems have part or all of To better understand what we mean by dynamic, Fig. 1 illus-
their input unknown and revealed dynamically during the execu- trates the route execution of a single vehicle D-VRP. Before the
tion of the routes, but in contrast with the latter category, exploit- vehicle leaves the depot (time t0), an initial route plans to visit
able stochastic knowledge is available on the dynamically revealed the currently known requests (A, B, C, D, E). While the vehicle exe-
information. As before, the vehicle routes can be redefined in an cutes its route, two new requests (X and Y) appear at time t1 and
ongoing fashion with the help of technological support. the initial route is adjusted to fulfill them. Finally, at time tf the
Besides dynamic routing problems, where customer visits must executed route is (A, B, C, D, Y, E, X).
be explicitly sequenced along the routes, there are other related This example reveals how dynamic routing inherently adjusts
vehicle dispatching problems, such as managing a fleet of emer- routes in an ongoing fashion, which requires real-time communi-
gency vehicles [23,54,66], or the so-called dynamic allocation prob- cation between vehicles and the dispatching center. Fig. 2 illus-
lems in the area of long haul truckload trucking [60,109,134]. In trates this real-time communication scheme, where the
this paper, we focus solely on dynamic problems with an explicit environment refers to the real-world while the dispatcher is the
routing dimension. agent that gives instructions to the vehicle. Once the vehicle is
The remainder of this document is organized as follows: Section 2 ready (first dotted arrow), the dispatcher makes a decision and in-
presents a general description of dynamic routing problems and structs the vehicle to fulfill request A (first double-headed arrow).
introduces the notion of degree of dynamism. Section 3 reviews dif- When the vehicle starts (second dotted arrow) and ends (third dot-
ferent applications in which dynamic routing problems arise, while ted arrow) service at request A, it notifies the dispatcher, which in
Section 4 provides a comprehensive survey of solution approaches. turn updates the available information and communicates the
Finally, Section 5 concludes this paper and gives directions for vehicle its next request (second double-headed arrow).
further research.
2.2. Differences with static routing
2. Dynamic vehicle routing problems
In contrast to their static counterparts, dynamic routing prob-
2.1. A general definition lems involve new elements that increase the complexity of their
decisions (more degrees of freedom) and introduce new challenges
The first reference to a dynamic vehicle routing problem is due while judging the merit of a given route plan.
to Wilson and Colvin [148]. They studied a single vehicle DARP, in In some contexts, such as the pick-up of express courier [52],
which customer requests are trips from an origin to a destination the transport company may deny a customer request. As a conse-
that appear dynamically. Their approach uses insertion heuristics quence, it can reject a request either because it is simply impossi-
able to perform well with low computational effort. Later, Psaraftis ble to service it, or because the cost of serving it is too high. This
[116] introduced the concept of immediate request: a customer process of acceptance/denial has been used in many approaches
requesting service always wants to be serviced as early as possible, [2,43,52,73–75,92] and is referred to as service guarantee [145].
requiring immediate replanning of the current vehicle route. In dynamic routing, the ability to redirect a moving vehicle to a
A number of technological advances have led to the multiplica- new request nearby allows for additional savings. Nevertheless, it
tion of real-time routing applications. With the introduction of the requires real-time knowledge of the vehicle position and being
Global Positioning System (GPS) in 1996, the development and able to communicate quickly with drivers to assign them new des-
widespread use of mobile and smart phones, combined with accu- tinations. Thus, this strategy has received limited interest, with the
rate Geographic Information Systems (GISs), companies are now main contributions being the early work by Regan et al. [120–122],
V. Pillac et al. / European Journal of Operational Research 225 (2013) 1–11 3
Different problems (or instances of a same problem) can have 3. A review of applications
different levels of dynamism, which can be characterized according
to two dimensions [76]: the frequency of changes and the urgency Recent advances in technology have allowed the emergence of a
of requests. The former is the rate at which new information be- wide new range of applications for vehicle routing. In particular,
comes available, while the latter is the time gap between the dis- the last decade has seen the development of Intelligent Transport
closure of a new request and its expected service time. From this Systems (ITSs), which are based on a combination of geolocation
observation three metrics have been proposed to measure the technologies, with precise geographic information systems, and
dynamism of a problem (or instance). increasingly efficient hardware and software for data processing
Lund et al. [95] defined the degree of dynamism d as the ratio be- and operations planning. We refer the interested reader to the
tween the number of dynamic requests nd and the total number of study by Crainic et al. [34] for more details on ITS and the contri-
requests ntot as follows: butions of operations research to this relatively new domain.
nd Among ITSs, the Advanced Fleet Management Systems (AFMSs)
d¼ ð1Þ are specifically designed for managing a corporate vehicle fleet.
ntot
The core problem is generally to deliver (pick-up) goods or persons
Based on the fact that the disclosure time of requests is also to (from) locations distributed in a given area. While customer re-
important [117,118], Larsen [87] proposed the effective degree of quests can either be known in advance or appear dynamically dur-
dynamism de. This metric can be interpreted as the normalized ing the day, vehicles are dispatched and routed in real time,
average of the disclosure times. Let T be the length of the planning potentially by taking into account changing traffic conditions,
4 V. Pillac et al. / European Journal of Operational Research 225 (2013) 1–11
uncertain demands, or varying service times. A key technological acquiring real-time traffic data to the dynamic routing of a fleet
feature of AFMSs is the optimization component. Traditionally, of vehicles. Zeimpekis et al. [153] proposed a Decision Support Sys-
vehicle routing relies on teams of human dispatchers, meaning a tem (DSS) for city logistics which takes into account dynamic tra-
critical operational process is bound to the competence and expe- vel and service times.
rience of dispatchers, as well as the management costs that are di- A typical application in city logistics is the courier service pres-
rectly linked to the size of the fleet [1]. Advances in computer ent in most urban areas. Couriers are dispatched to customer loca-
science have allowed a technological transfer from operational re- tions to collect packages, and either deliver them to their
search to AFMSs, as presented in the studies by Attanasio et al. [1], destination (short haul) or to a unique depot (long haul). Depend-
Du et al. [38], Godfrey and Powell [60], Powell and Topaloglu [115], ing on the level of service paid by the customer, couriers may con-
Roy [125], Simao et al. [129], and Slater [130]. solidate pick-ups from various customers, or provide an expedited
The remainder of this section presents applications where dy- service. Companies offering courier services often have an hetero-
namic routing has been or can be implemented. The interested geneous fleet composed of bicycles, motorbikes, cars, and small
reader is also referred to the work by Gendreau and Potvin [55] vans. The problem is then to dynamically route couriers, taking
and Ichoua et al. [76] for complementary reviews. into account not only the known requests, their type, pick-up
and delivery locations, and time windows, but also considering
3.1. Services traffic conditions and varying travel times. A case study by Attana-
sio et al. [1] outlines the benefits of using an optimization-enabled
In this category of applications, a service request is defined by a AFMS at eCourier Ltd, a London based company offering courier
customer location and a possible time window; while vehicle services. The authors illustrate that aside from the improvements
routes just fulfill service requests without considering side con- in service quality, response time, and courier efficiency, the use
straints such as capacity. Perhaps the simplest, yet most illustra- of an automated system allows decoupling the fleet size from the
tive case in this category is the dynamic traveling salesman need for more dispatchers. Further results motivated by a similar
problem [89]. application can be found in Gendreau et al. [51] and Ghiani et al.
A common application of dynamic routing can be found in the [59].
area of maintenance operations. Maintenance companies are often The delivery of newspapers and magazines is a domain in which
committed to their customers by means of a contract which spec- customer satisfaction is of first importance. When a magazine or
ifies periodical or planned visits to perform preventive mainte- newspaper is not delivered, a subscriber contacts a call center
nance, and may also request corrective maintenance on short and is offered to choose between a voucher or a future delivery.
notice. Therefore, each technician is first given a route with known In the latter case, the request is then forwarded to the delivery
requests at the beginning of the day, while new urgent requests are company, which assigns it to a driver that will do a priority deliv-
inserted dynamically throughout the day. An interesting feature of ery. Traditionally, this process relies on an exchange of phone calls,
this problem is the possible mix of skills, tools, and spare part faxes, and printed documents, that ultimately communicates the
requirements, which have to be matched in order to service the re- pending delivery to the driver once he/she comes back to the de-
quest. This problem has been studied by Borenstein et al. [19] with pot. As an alternative, Bieding et al. [18] propose a centralized
an application to British Telecom. application that makes use of mobile phones to communicate with
Another application of dynamic routing arises in the context of drivers and intelligently perform the routing in real time, reducing
the French non-profit organization SOS Médecins. This organization costs and improving customer satisfaction. More recently, Ferrucci
operates with a crew of physicians, who are called on duty via a et al. [44] developed an approach that makes use historical data to
call center coordinated with other emergency services. When a pa- anticipate future requests.
tient calls, the severity of the case is evaluated, and a visit by a Another application in which customer requests need to be an-
practitioner is planned accordingly. As in other emergency ser- swered with short delays can be found in companies with a direct
vices, having an efficient dispatching system reduces the response service model, such as grocery delivery services. In general, the
time, thus improving service level for the society. On the other customer selects products on a website, and then chooses a time
hand, it is important to decide in real-time whether or not to send frame for the delivery at his home. Traditionally, the vendor de-
a physician, so that it is possible to ensure a proper service level in fines an arbitrary number of customers that can be serviced within
areas where emergencies are likely to appear. a time window, and the time window is made unavailable to cus-
Dynamic aspects can also appear on arc routing problems. tomers as soon as the capacity is reached. Campbell and Savels-
This is for instance the case in the study by Tagmouti et al. bergh [24] defined the Home Delivery Problem, in which the goal
[136] on the operation of a fleet of vehicles for winter gritting is to maximize the total expected revenue by dynamically deciding
applications. Their work consider a network of streets or road whether or not to accept a customer request within a specific time
segments that need to be gritted when affected by a moving window. In comparison with the traditional approach, this means
storm. Depending on the movements of the storm, new segments that the time windows available for a customer are dynamically
may have to be gritted, and the routing of vehicles has to be up- defined taking into consideration the possible future requests.
dated accordingly. The authors propose a Greedy Randomized Adaptive Search Proce-
dure (GRASP) and compare different cost functions to capture the
3.2. Transport of goods problem uncertainty. Later, Azi et al. [3] proposed an Adaptive
Large Neighborhood Search (ALNS) that takes into account uncer-
Due to the fact that urban areas are often characterized by tainty by generating scenarios containing possible demand
highly variable traveling times, transport of goods in such areas realizations.
has led to the definition of a specific category of applications Apart from classical routing problems, related operational prob-
known as city logistics. City logistics can be defined as an integrated lems also arise in many organizations. The review by Stahlbock and
vision of transport activities in urban areas, taking into account Voss [135] on operations research applications in container termi-
factors such as traffic and competition or cooperation between nals describes the dynamic stacker crane problem [5,14], which
transport companies [140]. Barcelo et al. [6] developed a general considers the routing of container carriers loading and unloading
framework for city logistics applications. They describe the differ- ships in a terminal. Other applications include transport of
ent modules ranging from modeling the city road network and goods inside warehouses [132], factories, and hospitals, where
V. Pillac et al. / European Journal of Operational Research 225 (2013) 1–11 5
documents or expensive medical instruments must be transferred [78], Larsen et al. [91], and Zeimpekis et al. [154], to complement
efficiently between services [45]. our review.
The transport of persons is in general – and by many aspects – This section presents approaches that have been successfully
similar to the transport of goods, yet it is characterized by addi- applied to dynamic routing, in the absence of stochastic informa-
tional constraints such as regulation on waiting, travel, and service tion. In this context, critical information is revealed over time,
times. meaning that the complete instance is only known at the end of
Taxis are arguably the most common on-demand individual the planning horizon. As a consequence, exact methods only pro-
transport systems. Requests are composed of a pick-up location vide an optimal solution for the current state, but do not guarantee
and time, possibly coupled with a destination. They can be either that the solution will remain optimal once new data becomes
known in advance, for instance when a customer books a cab for available. Therefore, most dynamic approaches rely on heuristics
the next day, or they can arrive dynamically, in which case a taxi that quickly compute a solution to the current state of the problem.
must be dispatched in the shortest time. When customers cannot Approaches for dynamic and deterministic vehicle routing prob-
share a vehicle, the closest free taxi is generally the one which lems can be divided into two categories: those based on periodic
takes the ride, leaving limited space for optimization. The study reoptimization, and those based on continuous reoptimization.
by Caramia et al. [25], generalized by Fabri and Recht [42], focuses
on a multi-cab metropolitan transportation system, where a taxi 4.1.1. Periodic reoptimization
can transport more than one passenger at the same time. In this To the best of our knowledge, the first periodic reoptimization
case the online algorithms minimize the total traveled distance, approach is due to Psaraftis [116], with the development of a dy-
while assigning requests to vehicles and computing the taxi routes. namic programming approach. His research focuses on the DARP
This multi-cab transportation system can be generalized as an on- and consists in finding the optimal route each time a new request
demand or door-to-door transport service. is known. The main drawback of dynamic programming is the
Many applications involve the transport of children, the elderly, well-known curse of dimensionality [Chap. 1][110], which prevents
disabled people, or patients, from their home to schools, place of its application to large instances.
work, or medical centers. Xiang et al. [149] studied a DARP with More generally, periodic reoptimization approaches start at the
changing travel speeds, vehicle breakdowns, and traffic conges- beginning of the day with a first optimization that produces an ini-
tion; while Dial [36], followed by Horn [67–69], studied demand- tial set of routes. Then, an optimization procedure periodically
responsive transport systems. An extensive review of this class of solves a static problem corresponding to the current state, either
problems can be found in the studies by Cordeau et al. [32] and whenever the available data changes, or at fixed intervals of
Berbeglia et al. [14]. time – referred to as decision epochs [29] or time slices [81]. The
A singular application of on-demand transportation systems advantage of periodic reoptimization is that it can be based on
can be found in major hospitals, with services possibly spread algorithms developed for static routing, for which extensive
across various buildings on several branches. Depending on the research has been carried out. The main drawback is that all the
medical procedure or facility capacity, a patient may need to be optimization needs to be performed before updating the routing
transferred on short notice from one service to another, possibly plan, thus increasing delays for the dispatcher.
requiring trained staff or specific equipment for his/her care. This Yang et al. [150] addressed the real-time truckload PDP, in
application has been studied by Beaudry et al. [7], Kergosien which a fleet of trucks has to service point-to-point transport re-
et al. [80], and Melachrinoudis et al. [96]. quests arriving dynamically. Important assumptions are that all
Air taxis developed as a flexible response to the limitations of trucks can only handle one request at a time, with no possible pre-
traditional airlines. Air taxis offer passengers the opportunity to emption, and they travel at the same constant speed. The authors
travel through smaller airports, avoiding waiting lines at check-in propose MYOPT, a rolling horizon approach based on a linear pro-
and security checks. Air taxi companies offer an on-demand ser- gram (LP) that is solved whenever a new request arrives. Along the
vice: customers book a flight a few days in advance, specifying same line of linear programming, Chen and Xu [29] designed a dy-
whether they are willing to share the aircraft, stop at an interme- namic column generation algorithm (DYCOL) for the D-VRPTW.
diate airport, or have flexible traveling hours. Then, the company The authors propose the concept of decision epochs over the plan-
accommodates these requests, trying to consolidate flights when- ning horizon, which are the dates when the optimization process
ever possible. The underlying optimization problems have not runs. The novelty of their approach relies on dynamically generat-
been subject to much attention, except in the studies by Cordeau ing columns for a set-partitioning model, using columns from the
et al. [32], Espinoza et al. [40,41], Fagerholt et al. [43], and Yao previous decision epoch. The authors compared DYCOL to a tradi-
et al. [152]. Similar problems arises in helicopter transportation tional column generation with no time limit (COL). Computational
systems, typically used by oil and gas companies to transport per- results based on the Solomon benchmark [133] demonstrate that
sonnel between offshore petroleum platforms [63,124]. DYCOL yields comparable results in terms of objective function,
but with running times limited to 10 seconds, opposed to the
various hours consumed by COL.
4. Solution methods Montemanni et al. [102] developed an Ant Colony System (ACS)
to solve the D-VRP. Similar to Kilby et al. [81], their approach uses
Few research was conducted on dynamic routing between the time slices, that is, they divide the day in periods of equal duration.
work of Psaraftis [116] in 1980 and the late 1990s. However, the A request arriving during a time slice is not handled until the end
last decade has seen a renewed interest for this class of problems of the time bucket, thus the problem solved during a time slice
[39], with solution techniques ranging from linear programming only considers the requests known at its beginning. Hence, the
to metaheuristics. This section presents the major contributions optimization is run statically and independently during each time
in this field, and the reader is referred to the reviews, books, and slice. The main advantage of this time partition is that similar com-
special issues by Gendreau and Potvin [55,56], Ghiani et al. [57], putational effort is allowed for each time slice. This discretization
Goel [61], Ichoua [72], Ichoua et al. [75,76], Jaillet and Wagner is also possible by the nature of the requests, which are never
6 V. Pillac et al. / European Journal of Operational Research 225 (2013) 1–11
urgent, and can be postponed. An interesting feature of their ap- of scenarios to accurately reflect reality. Alternatively, stochastic
proach is the use of the pheromone trace to transfer characteristics modeling strategies formally capture the stochastic nature of the
of a good solution to the next time slice. A similar approach was problem, but they are highly technical in their formulation and re-
also used by Gambardella et al. [50] and Rizzoli et al. [123]. quire to efficiently compute possibly complex expected values.
Examples of these two strategies follow.
4.1.2. Continuous reoptimization
Continuous reoptimization approaches perform the optimiza- 4.2.1. Stochastic modeling
tion throughout the day and maintain information on good solu- Powell et al. [113] formulated a truckload PDP as a Markov
tions in an adaptive memory [138]. Whenever the available data Decision Process (MDP). Later, MDPs were used by Thomas and
changes, a decision procedure aggregates the information from White [142] and Thomas [141] to solve a VRP in which known cus-
the memory to update the current routing. The advantage is that tomers may ask for service with a known probability. Kim et al.
the computational capacity is maximized, possibly at the expense [82] also used MDPs to tackle the VRP with dynamic travel times.
of a more complex implementation. It is worth noting that because Unfortunately, the curse of dimensionality and the simplifying
the current routing is subject to change at any time, vehicles do not assumptions make this approach unsuitable in most real-world
know their next destination until they finish the service of a applications. Nonetheless, it allowed new insights in the field of
request. dynamic programming.
To the best of our knowledge, the first continuous reoptimiza- To cope with the scalability problems of traditional dynamic
tion approach is due to Gendreau et al. [52] with the adaptation programming, Approximate Dynamic Programming (ADP) steps
of the parallel Tabu Search (TS) framework introduced by Taillard forward in time, approximates the value function, and ultimately
et al. [137] to a D-VRPTW problem arising in the local operation avoids the evaluation of all possible states. We refer the interested
of long distance express courier services. Their approach maintains reader to Powell [110,111] for a more detailed description of the
a pool of good routes – the adaptive memory – which is used to ADP framework. ADP has been successfully applied to freight
generate initial solutions for a parallel TS. The parallelized search transport [112,114] and fleet management problems
is done by partitioning the routes of the current solution, and opti- [60,115,129]. In particular, Novoa and Storer [104] propose an
mizing them in independent threads. Whenever a new customer ADP algorithm to dynamically solve the VRPSD.
request arrives, it is checked against all the solutions from the Linear programming has also been adapted to the dynamic and
adaptive memory to decide whether it should be accepted or re- stochastic context. The OPTUN approach, proposed by Yang et al.
jected. This framework was also implemented for the D-VRP [150] as an extension of MYOPT (see Section 4.1.1), considers
[73,74], while other variations of TS have been applied to the D- opportunity costs on each arc to reflect the expected cost of travel-
PDP [6,27] and the DARP [2,7]. ing to isolated areas. Consequently, the optimization tends to reject
Bent and Van Hentenryck [9] introduced the Multiple Plan Ap- isolated requests, and avoids traversing arcs that are far away from
proach (MPA) as a generalization of the TS with adaptive memory potential requests. Later, Yang et al. [151] studied the emergency
[52]. The general idea is to populate and maintain a solution pool vehicle dispatching and routing and proposed a mathematical for-
(the routing plans) that are used to generate a distinguished solu- mulation that was later used by Haghani and Yang [66] on a similar
tion. Whenever a new request arrives, a procedure is called to problem.
check whether it can be serviced or not; if it can be serviced, then
the request is inserted in the solution pool and incompatible solu- 4.2.2. Sampling
tions are discarded. Pool updates are performed periodically or Sampling approaches rely on the generation of scenarios con-
whenever a vehicle finishes servicing a customer. This pool-update taining possible realizations of the random variables. Fig. 3 illus-
phase is crucial and ensures that all solutions are coherent with the trates how scenarios are generated for the D-VRP. Solely based
current state of vehicles and customers. The pool can be seen as an on the current customers (Fig. 3a), the optimal tour would be
adaptive memory that maintains a set of alternative solutions. (A, B, E, D, C), which ignores two zones (gray areas) where customers
In an early work, Benyahia and Potvin [13] studied the D-PDP are likely to appear. By sampling the customer spatial distributions
and proposed a Genetic Algorithm (GA) that models the decision (Fig. 3b), customers X, Y, and Z are generated, and the new optimal
process of a human dispatcher. More recently, GAs were also used tour is (C, X, Y, B, A, Z, E, D). Removing the sampled (potential) cus-
for the same problem [30,65] and for the D-VRP [144]. Genetic tomers (Fig. 3c) leads to the tour (C, B, A, E, D) which is suboptimal
algorithms in dynamic contexts are very similar to those designed regarding a myopic cost evaluation, but leaves room to accommo-
for static problems, although they generally run throughout the date new customers at a lower cost.
planning horizon and solutions are constantly adapting to the The Multiple Scenario Approach (MSA) is a predictive adapta-
changes made to the input. tion of the MPA framework discussed in Section 4.1.2. The idea be-
hind MSA is to take advantage of the time between decisions to
4.2. Dynamic and stochastic routing problems continuously improve the current scenario pool. During the initial-
ization, the algorithm, generates a first set of scenarios based on
Dynamic and stochastic routing problems can be seen as an the requests known beforehand. Throughout the day, scenarios
extension of their deterministic counterparts, where additional are then reoptimized and new ones are generated and added to
(stochastic) knowledge is available in the dynamically revealed in- the pool. When a decision is required, the scenario optimization
put. Approaches for this class of problems can be divided in two procedure is suspended, and MSA uses the scenario pool to select
categories: those based on sampling and those based on stochastic the request to service next. MSA then discards the scenarios that
modeling. As their name suggests, sampling strategies incorporate are incompatible with the current routing, and resumes the opti-
stochastic knowledge by generating scenarios based on realizations mization. Computational experiments on instances adapted from
drawn from the random variable distributions. Each scenario is the Solomon benchmark [133] showed that MSA outperforms
then optimized by solving the static and deterministic problem MPA both in terms of serviced customers and traveled distances,
they define. On the other hand, approaches based on stochastic especially for instances with high degrees of dynamism [9]. Flat-
modeling integrate stochastic knowledge analytically. The advan- berg et al. [47] adapted the SPIDER commercial solver to use multi-
tage of sampling is its relative simplicity and flexibility on distribu- ple scenarios and a consensus algorithm to tackle the D-VRP, while
tional assumptions, while its drawback is the massive generation Pillac et al. [107] implemented an event-driven optimization
V. Pillac et al. / European Journal of Operational Research 225 (2013) 1–11 7
framework based on MSA and showed significant improvements windows, where time lags appear between requests. Mitrović-
over state-of-the-art algorithms for the D-VRPSD. Minić [100] proved that in all cases it is better to wait after servic-
An important component of scenario based-approaches such as ing a customer, but a more refined strategy can lead to further
MSA is the decision process, which defines how the information improvements. The problem is in general to evaluate the likelihood
from the scenario pool is used to reach upon a decision regarding of a new request in the neighborhood of a serviced request and to
the next customer to visit. The most common algorithms used to plan a waiting period accordingly. The waiting strategy has been
reach a decision in MSA are: consensus, expectation, and regret. implemented in various frameworks for the D-VRP [22,141], D-
The consensus algorithm [9,10] selects the customer appearing VRPTW [12,21,75,145], D-PDP [100,59], and Dynamic and Stochas-
first with the highest frequency among scenarios. Expectation tic TSP [58]. The strategy has shown good results, especially in the
[8,10,26] consists in evaluating the cost of visiting each customer case of a limited fleet facing a high request rate [145].
first by forcing its visit in all scenarios and performing a complete Aside from the waiting after or before servicing a customer, a
optimization. Finally, regret [8] approximates the expectation algo- vehicle can be relocated to a strategic position, where new
rithm and avoids the reoptimization of all scenarios. Even though requests are likely to appear. This strategy is the keystone of
these algorithms were initially designed for the routing of a single emergency fleet deployment, also known as Emergency Vehicle
vehicle, they can be extended to the multi-vehicle case [145]. Dispatching – or Redeployment – Problem [54,66]. The reloca-
Hvattum et al. [70] developed the Dynamic Sample Scenario tion strategy has also been applied to other vehicle routing prob-
Hedge Heuristic (DSHH), an approach similar to the consensus lems, such as the D-VRP [87], D-VRPTW [12,21,75,145], D-TSPTW
algorithm for D-VRP. This method divides the planning horizon [89], D-PDP [59,119], and the Resource Allocation Problem (RAP)
into time intervals. At the beginning of each interval, DSHH revises [60].
the routing by assigning a subset of promising requests to the vehi- Request buffering, introduced by Pureza and Laporte [119], con-
cles, depending on the frequency of their assignment over all sce- sists in delaying the assignment of some requests to vehicles in a
narios. DSHH later led to the development of the Branch and Regret priority buffer, so that more urgent requests can be handled first.
Heuristic (BRH), where scenarios are merged to build a unique
solution. 4.3. Performance evaluation
Various local search approaches have been developed for the
stochastic and dynamic problems. Ghiani et al. [59] developed an In contrast to static problems, where measuring the perfor-
algorithm for the D-PDP that only samples the near future to re- mance of an algorithm is straightforward (i.e., running time and
duce the computational effort. The main difference with MSA is solution quality), dynamic problems require the introduction of
that no scenario pool is used and the selection of the distinguished new metrics to assess the performance of a particular method.
solution is based on the expected penalty of accommodating re- Sleator and Tarjan [131] introduced the competitive analysis
quests in the near future. Azi et al. [3] developed an Adaptive Large [77,90]. Let P be a minimization problem and I the set of all in-
Neighborhood Search (ALNS) for a dynamic routing problem with stances of P. Let z⁄(Ioff) be the optimal cost for the offline instance
multiple delivery routes, in which the dynamic decision is the Ioff corresponding to I 2 I . For offline instance Ioff, all input data
acceptance of a new request. The approach maintains a pool of sce- from instance I, either static or dynamic, is available when building
narios, optimized by an ALNS, that are used to evaluate the oppor- the solution. In contrast, the data of the online version I is revealed
tunity value of an incoming request. in real time, thus an algorithm A has to take into account new
Tabu search has also been adapted to dynamic and stochastic information as it is revealed and produce a solution relevant to
problems. Ichoua et al. [75] and Attanasio et al. [1] tackled with the current state of knowledge. Let zA ðIÞ ¼ zðxA ðIÞÞ be the cost of
tabu search the D-VRPTW and the D-PDP, respectively. the final solution xA ðIÞ found by the online algorithm A on instance
I. Algorithm A is said to be c-competitive, or equivalently to have a
4.2.3. Other strategies competitive ratio of c, if there exists a constant a such that
In addition to the general frameworks described previously, the
zA ðIÞ 6 c z ðIoff Þ þ a; 8I 2 I ð4Þ
use of stochastic knowledge allows for the design and implemen-
tation of other strategies that try to adequately respond to upcom- In the case where a = 0, the algorithm is said to be strictly c-compet-
ing events. itive, meaning that in all cases the objective value of the solution
The waiting strategy consists in deciding whether a vehicle found by A will be at most of c times the optimal value. The com-
should wait after servicing a request, before heading toward the petitive ratio metric allows a worst-case absolute measure of an
next customer; or planning a waiting period on a strategic location. algorithm performance in terms of the objective value. We refer
This strategy is particularly important in problems with time the reader to Borodin and El-Yaniv [20] for an in-depth analysis of
8 V. Pillac et al. / European Journal of Operational Research 225 (2013) 1–11
this measure, and to Jaillet and Wagner [77] and Fink et al. [46] for considered distributed optimization [52], most approaches re-
results on various routing problems. viewed in this document do not take advantage of parallel archi-
The main drawback of the competitive analysis is that it re- tectures. The development of parallel algorithms is a challenge
quires to prove the previously stated inequality analytically, which that could reduce the time needed for optimization and provide
may be complex for real-world applications. The value of informa- decision makers with highly reactive tools.
tion proposed by Mitrović-Minić [100] constitutes a more flexible Fourth, our review of the existing literature revealed that a large
and practical metric. We denote by zA ðIoff Þ the value of the objec- fraction of work done in the area of dynamic routing does not con-
tive function returned by algorithm A for the offline instance Ioff. sider stochastic aspects. We are convinced that developing algo-
The value of information V A ðIÞ for algorithm A on instance I is then rithms that make use of stochastic information will improve the
defined as fleet performance and reduce operating costs. Thus this line of re-
search should become a priority in the near future.
zA ðIÞ zA ðIoff Þ
V A ðIÞ ¼ ð5Þ Finally, researchers have mainly focused on the routing aspect
zA ðIoff Þ
of the dynamic fleet management. However, in some applications
The value of information can be interpreted as the gap between the there is more that can be done to improve performance and service
solution returned by an algorithm A on an instance I and the solu- level. For instance, in equipment maintenance services, the call
tion returned by the same algorithm when all information from I is center has a certain degree of freedom in fixing service appoint-
known beforehand. In contrast with the competitive ratio, the value ments. In other words, it means that the customer time windows
of information gives information on the performance of an algo- can be defined, or influenced, by the call center operator. As a con-
rithm based on empirical results, without requiring optimal solu- sequence, a system in which aside from giving a yes/no answer to a
tions for the offline instances. It captures the impact of the customer request, suggests convenient times for the company
dynamism on the solution yield by the algorithm under analysis. would be highly desirable in such contexts.
For instance, Gendreau et al. [52] report a value of information be-
tween 2.5% and 4.1% for their tabu search algorithm for the D- Acknowledgements
VRPTW, while Tagmouti et al. [136] report values between 10%
and 26.7% for a variable neighborhood search descent applied to a Financial support for this work was provided by the CPER (Con-
dynamic arc routing problem. trat de Projet Etat Region) Vallée du Libre; and the Centro de Estu-
dios Interdisciplinarios Básicos y Aplicados en Complejidad (CEIBA,
4.4. Benchmarks Colombia). This support is gratefully acknowledged.
[15] D. Bertsimas, Probabilistic combinatorial optimization problems, PhD thesis, [44] F. Ferrucci, S. Block, M. Gendreau, A pro-active real-time control approach for
Massachusetts Institute of Technology, Dept. of Mathematics, 1988. dynamic vehicle routing problems dealing with the delivery of urgent goods,
[16] D. Bertsimas, D. Simchi-Levi, A new generation of vehicle routing research: European Journal of Operational Research 225 (1) (2013) 130–141.
robust algorithms, addressing uncertainty, Operations Research 44 (2) (1996) [45] C. Fiegl, C. Pontow, Online scheduling of pick-up and delivery tasks in
286–304. hospitals, Journal of Biomedical Informatics 42 (4) (2009) 624–632.
[17] D. Bertsimas, G. Van Ryzin, A stochastic and dynamic vehicle-routing problem [46] I. Fink, S.O. Krumke, S. Westphal, New lower bounds for online k-server
in the Euclidean plane, Operations Research 39 (4) (1991) 601–615. routing problems, Information Processing Letters 109 (11) (2009) 563–567.
[18] T. Bieding, S. Görtz, A. Klose, On-line routing per mobile phone: a case on [47] T. Flatberg, G. Hasle, O. Kloster, E.J. Nilssen, A. Riise, Dynamic and stochastic
subsequent deliveries of newspapers, in: M. Beckmann, H.P. Künzi, G. Fandel, vehicle routing in practice, in: V. Zeimpekis, C.D. Tarantilis, G.M. Giaglis, I.
W. Trockel, A. Basile, A. Drexl, H. Dawid, K. Inderfurth, W. Kürsten, J.A. Minis (Eds.), Dynamic Fleet Management, Operations Research/Computer
Nunen, M.G. Speranza, L. Bertazzi (Eds.), Innovations in Distribution Logistics, Science Interfaces, vol. 38, Springer, US, 2007, pp. 41–63.
Lecture Notes in Economics and Mathematical Systems, vol. 619, Springer, [48] B. Fleischmann, S. Gnutzmann, E. Sandvoss, Dynamic vehicle routing based on
Berlin Heidelberg, 2009, pp. 29–51. online traffic information, Transportation Science 38 (4) (2004) 420–433.
[19] Y. Borenstein, N. Shah, E. Tsang, R. Dorne, A. Alsheddy, C. Voudouris, On the [49] M. Flood, The traveling-salesman problem, Operations Research 4 (1) (1956)
partitioning of dynamic workforce scheduling problems, Journal of 61–75.
Scheduling 13 (4) (2010) 411–425. [50] L. Gambardella, A. Rizzoli, F. Oliverio, N. Casagrande, A. Donati, R.
[20] A. Borodin, R. El-Yaniv, Online Computation and Competitive Analysis, Montemanni, E. Lucibello, Ant colony optimization for vehicle routing in
Cambridge University Press, Cambridge, 2005. advanced logistics systems, in: Proceedings of the International Workshop on
[21] R.M. Branchini, V.A. Armentano, A. Lokketangen, Adaptive granular local Modelling and Applied Simulation (MAS 2003), 2003, pp. 3–9.
search heuristic for a dynamic vehicle routing problem, Computers & [51] M. Gendreau, F. Guertin, J.-Y. Potvin, R. Séguin, Neighborhood search
Operations Research 36 (11) (2009) 2955–2968. heuristics for a dynamic vehicle dispatching problem with pick-ups and
[22] J. Branke, M. Middendorf, G. Noeth, M. Dessouky, Waiting strategies for deliveries, Transportation Research Part C: Emerging Technologies 14 (3)
dynamic vehicle routing, Transportation Science 39 (3) (2005) 298–312. (2006) 157–174.
[23] L. Brotcorne, G. Laporte, F. Semet, Ambulance location and relocation models, [52] M. Gendreau, F. Guertin, J.-Y. Potvin, E. Taillard, Parallel tabu search for real-
European Journal of Operational Research 147 (3) (2003) 451–463. time vehicle routing and dispatching, Transportation Science 33 (4) (1999)
[24] A. Campbell, M. Savelsbergh, Decision support for consumer direct grocery 381–390.
initiatives, Transportation Science 39 (3) (2005) 313–327. [53] M. Gendreau, G. Laporte, R. Séguin, Stochastic vehicle routing, European
[25] M. Caramia, G. Italiano, G. Oriolo, A. Pacifici, A. Perugia, Routing a fleet of Journal of Operational Research 88 (1) (1996) 3–12.
vehicles for dynamic combined pick-up and deliveries services, in: [54] M. Gendreau, G. Laporte, F. Semet, A dynamic model and parallel tabu search
Proceedings of the Symposium on Operation Research 2001, Duisburg, heuristic for real-time ambulance relocation, Parallel Computing 27 (12)
Germany, 2002, pp. 3–5. (2001) 1641–1653.
[26] H. Chang, R. Givan, E. Chong, On-line scheduling via sampling, in: Proceedings [55] M. Gendreau, J.-Y. Potvin, Dynamic vehicle routing and dispatching, in:
of the Artificial Intelligence Planning and Scheduling (AIPS 2000), 2000, pp. Teodor G. Crainic, Gilbert Laporte (Eds.), Fleet Management and Logistics,
62–71. Kluwer, Boston, 1998, pp. 115–126 (chapter 5).
[27] M.S. Chang, S. Chen, C. Hsueh, Real-time vehicle routing problem with time [56] M. Gendreau, J.-Y. Potvin (Eds.) Transportation Science. Number 38. (special
windows and simultaneous delivery/pickup demands, Journal of the Eastern issue on real-time fleet management), 2004.
Asia Society for Transportation Studies 5 (2003) 2273–2286. [57] G. Ghiani, F. Guerriero, G. Laporte, R. Musmanno, vehicle routing: solution
[28] H.-K. Chen, C.-F. Hsueh, M.-S. Chang, The real-time time-dependent vehicle concepts, algorithms and parallel computing strategies, European Journal of
routing problem, Transportation Research Part E: Logistics and Operational Research 151 (1) (2003) 1–11.
Transportation Review 42 (5) (2006) 383–408. [58] G. Ghiani, G. Laporte, E. Manni, R. Musmanno, Waiting strategies for the
[29] Z. Chen, H. Xu, Dynamic column generation for dynamic vehicle routing with dynamic and stochastic traveling salesman problem, Int J Operat Res 5 (4)
time windows, Transportation Science 40 (1) (2006) 74–88. (2008) 233–241.
[30] B.K.S. Cheung, K.L. Choy, C.-L. Li, W. Shi, J. Tang, Dynamic routing model and [59] G. Ghiani, E. Manni, A. Quaranta, C. Triki, Anticipatory algorithms for same-
solution methods for fleet management with mobile technologies, day courier dispatching, Transportation Research Part E: Logistics and
International Journal of Production Economics 113 (2) (2008) 694–705. Transportation Review 45 (1) (2009) 96–106.
[31] C. Christiansen, J. Lysgaard, A branch-and-price algorithm for the capacitated [60] G. Godfrey, W.B. Powell, An adaptive dynamic programming algorithm for
vehicle routing problem with stochastic demands, Operations Research dynamic fleet management, I: single period travel times, Transportation
Letters 35 (6) (2007) 773–781. Science 36 (1) (2002) 21–39.
[32] J.-F. Cordeau, G. Laporte, J.-Y. Potvin, M.W. Savelsbergh, Transportation on [61] A. Goel, in: Fleet Telematics: Real-Time Management and Planning of
demand, in: C. Barnhart, G. Laporte (Eds.), Transportation, Handbooks in Commercial Vehicle Operations, Operations Research Computer Science
Operations Research and Management Science, vol. 14, Elsevier, 2007, pp. Interfaces Series, vol. 40, Springer Verlag, 2008.
429–466 (chapter 7). [62] A. Goel, V. Gruhn, A general vehicle routing problem, European Journal of
[33] J.-F. Cordeau, G. Laporte, M.W. Savelsbergh, D. Vigo, Vehicle routing, in: C. Operational Research 191 (3) (2008) 650–660.
Barnhart, G. Laporte (Eds.), Transportation, Handbooks in Operations [63] I. Gribkovskaia, G. Laporte, A. Shlopak, A tabu search heuristic for a routing
Research and Management Science, vol. 14, Elsevier, 2007, pp. 367–428 problem arising in servicing of offshore oil and gas platforms, Journal of the
(chapter 6). Operational Research Society 59 (11) (2008) 1449–1459.
[34] T.G. Crainic, M. Gendreau, J.-Y. Potvin, Intelligent freight-transportation [64] A.R. Güner, A. Murat, R.B. Chinnam, Dynamic routing under recurrent and
systems: assessment and the contribution of operations research, non-recurrent congestion using real-time its information, Computers &
Transportation Research Part C: Emerging Technologies 17 (6) (2009) 541– Operations Research 39 (2) (2012) 358–373.
557. [65] A. Haghani, S. Jung, A dynamic vehicle routing problem with time-dependent
[35] G. Dantzig, J. Ramser, The truck dispatching problem, Management Science 6 travel times, Computers & Operations Research 32 (11) (2005) 2959–2986.
(1) (1959) 80–91. [66] A. Haghani, S. Yang, Real-time emergency response fleet deployment:
[36] R.B. Dial, Autonomous dial-a-ride transit introductory overview, concepts, systems, simulation & case studies, in: V. Zeimpekis, C.D.
Transportation Research Part C: Emerging Technologies 3 (5) (1995) 261– Tarantilis, G.M. Giaglis, I. Minis (Eds.), Dynamic Fleet Management,
275. Operations Research/Computer Science Interfaces, vol. 38, Springer, US,
[37] M. Dror, G. Laporte, P. Trudeau, Vehicle routing with stochastic demands: 2007, pp. 133–162.
Properties and solution frameworks, Transportation Science 23 (3) (1989) [67] M.E.T. Horn, Fleet scheduling and dispatching for demand-responsive
166–176. passenger services, Transportation Research Part C: Emerging Technologies
[38] T. Du, F.K. Wang, P.-Y. Lu, A real-time vehicle-dispatching system for 10 (1) (2002) 35–63.
consolidating milk runs, Transportation Research Part E: Logistics and [68] M.E.T. Horn, Multi-modal and demand-responsive passenger transport
Transportation Review 43 (5) (2007) 565–577. systems: a modelling framework with embedded control systems,
[39] B. Eksioglu, A.V. Vural, A. Reisman, The vehicle routing problem: a taxonomic Transportation Research Part A: Policy and Practice 36 (2) (2002) 167–
review, Computers & Industrial Engineering 57 (4) (2009) 1472–1483. 188.
[40] D. Espinoza, R. Garcia, M. Goycoolea, G.L. Nemhauser, M.W.P. Savelsbergh, [69] M.E.T. Horn, Procedures for planning multi-leg journeys with fixed-route and
Per-seat, on-demand air transportation part I: problem description and an demand-responsive passenger transport services, Transportation Research
integer multicommodity flow model, Transportation Science 42 (3) (2008) Part C: Emerging Technologies 12 (1) (2004) 33–55.
263–278. [70] L.M. Hvattum, A. Lokketangen, G. Laporte, Solving a dynamic and stochastic
[41] D. Espinoza, R. Garcia, M. Goycoolea, G.L. Nemhauser, M.W.P. Savelsbergh, vehicle routing problem with a sample scenario hedging heuristic,
Per-seat, on-demand air transportation part II: parallel local search, Transportation Science 40 (4) (2006) 421–438.
Transportation Science 42 (3) (2008) 279–291. [71] L.M. Hvattum, A. Lokketangen, G. Laporte, A branch-and-regret heuristic for
[42] A. Fabri, P. Recht, On dynamic pickup and delivery vehicle routing with stochastic and dynamic vehicle routing problems, Networks 49 (4) (2007)
several time windows and waiting times, Transportation Research Part B: 330–340.
Methodological 40 (4) (2006) 335–350. [72] S. Ichoua, Problèmes de gestion de flottes de véhicules en temps réel, PhD
[43] K. Fagerholt, B.A. Foss, O.J. Horgen, A decision support model for establishing thesis, Universite de Montreal, Montreal, Canada, 2001.
an air taxi service: a case study, Journal of The Operational Research Society [73] S. Ichoua, M. Gendreau, J.-Y. Potvin, Diversion issues in real-time vehicle
60 (9) (2009) 1173–1182. dispatching, Transportation Science 34 (4) (2000) 426–438.
10 V. Pillac et al. / European Journal of Operational Research 225 (2013) 1–11
[74] S. Ichoua, M. Gendreau, J.-Y. Potvin, Vehicle dispatching with time-dependent [102] R. Montemanni, L.M. Gambardella, A.E. Rizzoli, A.V. Donati, Ant colony
travel times, European Journal of Operational Research 144 (2) (2003) 379– system for a dynamic vehicle routing problem, Journal of Combinatorial
396. Optimization 10 (4) (2005) 327–343.
[75] S. Ichoua, M. Gendreau, J.-Y. Potvin, Exploiting knowledge about future [103] Q. Mu, Z. Fu, J. Lysgaard, R. Eglese, Disruption management of the vehicle
demands for real-time vehicle dispatching, Transportation Science 40 (2) routing problem with vehicle breakdown, Journal of the Operational
(2006) 211–225. Research Society 62 (4) (2011) 742–749.
[76] S. Ichoua, M. Gendreau, J.-Y. Potvin, Planned route optimization for real-time [104] C. Novoa, R. Storer, An approximate dynamic programming approach for the
vehicle routing, in: V. Zeimpekis, C.D. Tarantilis, G.M. Giaglis, I. Minis (Eds.), vehicle routing problem with stochastic demands, European Journal of
Dynamic Fleet Management, Operations Research/Computer Science Operational Research 196 (2) (2009) 509–515.
Interfaces, vol. 38, Springer, US, 2007, pp. 1–18. [105] C.M. Novoa, Static and dynamic approaches for solving the vehicle routing
[77] P. Jaillet, M.R. Wagner, Generalized online routing: new competitive ratios, problem with stochastic demands, PhD thesis, Lehigh University,
resource augmentation, and asymptotic analyses, Operations Research 56 (3) Pennsylvania, United States. AAT 3188502, 2005.
(2008) 745–757. [106] G. Pankratz, V. Krypczyk, Benchmark data sets for dynamic vehicle routing
[78] P. Jaillet, M.R. Wagner, Online vehicle routing problems: a survey, in: The problems <http://www.fernuni-hagen.de/WINF/inhalte/benchmark_data.
Vehicle Routing Problem: Latest Advances and New Challenges, Operations htm>, 2009.
Research/Computer Science Interfaces Series, vol. 43, Springer, US, 2008, pp. [107] V. Pillac, C. Guéret, A.L. Medaglia, An event-driven optimization framework
221–237. for dynamic vehicle routing, Decision Support Systems, in press, http://
[79] A.S. Kenyon, D. Morton, Stochastic vehicle routing with random travel times, dx.doi.org/10.1016/j.dss.2012.06.007.
Transportation Science 37 (1) (2003) 69. [108] J.Y. Potvin, Y. Xu, I. Benyahia, Vehicle routing and scheduling with dynamic
[80] Y. Kergosien, C. Lenté, D. Piton, J.-C. Billaut, A tabu search heuristic for the travel times, Computers & Operations Research 33 (4) (2006) 1129–1137.
dynamic transportation of patients between care units, European Journal of [109] W. Powell, J. Shapiro, H. Simao, An adaptive dynamic programming algorithm
Operational Research 214 (2) (2011) 442–452. for the heterogeneous resource allocation problem, Transportation Science
[81] P. Kilby, P. Prosser, P. Shaw, Dynamic VRPs: a study of scenarios, Technical 36 (2) (2002) 231–249.
Report APES-06-1998, University of Strathclyde, Glasgow, Scotland, 1998. [110] W.B. Powell, Approximate dynamic programming: solving the curses of
[82] S. Kim, M. Lewis, C.C. W, Optimal vehicle routing with real-time traffic dimensionality, Wiley Series in Probability and Statistics, vol. 703, Wiley-
information, IEEE Transactions on Intelligent Transportation Systems 6 (2) Interscience, Hoboken, New Jersey, 2007.
(2005) 178–188. [111] W.B. Powell, What you should know about approximate dynamic
[83] G. Laporte, What you should know about the vehicle routing problem, Naval programming, Naval Research Logistics 56 (3) (2009) 239–249.
Research Logistics 54 (8) (2007) 811–819. [112] W.B. Powell, B. Bouzaiene-Ayari, H. Simao, Dynamic models for freight
[84] G. Laporte, Fifty years of vehicle routing, Transportation Science 43 (4) (2009) transportation, in: C. Barnhart, G. Laporte (Eds.), Transportation, Handbooks
408–416. in Operations Research and Management Science, vol. 14, North-Holland,
[85] G. Laporte, F. Louveaux, H. Mercure, The vehicle routing problem with 2007, pp. 285–365 (chapter 5).
stochastic travel times, Transportation Science 26 (3) (1992) 161–170. [113] W.B. Powell, Y. Sheffi, K.S. Nickerson, K. Butterbaugh, S. Atherton, Maximizing
[86] G. Laporte, F. Louveaux, L. Van Hamme, An integer L-shaped algorithm for the profits for North American Van Lines’ truckload division: a new framework
capacitated vehicle routing problem with stochastic demands, Operations for pricing and operation, Interfaces 18 (1) (1988) 21–41.
Research 50 (3) (2002) 415–423. [114] W.B. Powell, H. Topaloglu, Stochastic programming in transportation and
[87] A. Larsen, The Dynamic Vehicle Routing Problem, PhD thesis, Technical logistics, Handbooks in Operations Research and Management Science 10
University of Denmark (DTU), 2001. (2003) 555–636.
[88] A. Larsen, O.B.G. Madsen, M.M. Solomon, Partially dynamic vehicle routing – [115] W.B. Powell, H. Topaloglu, Fleet management, in: S. Wallace, W. Ziemba
models and algorithms, The Journal of the Operational Research Society 53 (Eds.), Applications of Stochastic Programming, MPS-SIAM Series on
(6) (2002) 637–646. Optimization, vol. 5, SIAM, 2005, pp. 185–215 (chapter 12).
[89] A. Larsen, O.B.G. Madsen, M.M. Solomon, The a priori dynamic traveling [116] H. Psaraftis, A dynamic-programming solution to the single vehicle many-to-
salesman problem with time windows, Transportation Science 38 (4) (2004) many immediate request dial-a-ride problem, Transportation Science 14 (2)
459–472. (1980) 130–154.
[90] A. Larsen, O.B.G. Madsen, M.M. Solomon, Classification of dynamic vehicle [117] H. Psaraftis, Dynamic vehicle routing problems, in: B. Golden, A. Assas (Eds.),
routing systems, in: V. Zeimpekis, C.D. Tarantilis, G.M. Giaglis, I. Minis (Eds.), Vehicle Routing: Methods and Studies, Elsevier Science Publishers BV, 1988,
Dynamic Fleet Management, Operations Research/Computer Science pp. 223–248.
Interfaces Series, vol. 38, Springer, US, 2007, pp. 19–40 (chapter 2). [118] H.N. Psaraftis, Dynamic vehicle routing: status and prospects, Annals of
[91] A. Larsen, O.B.G. Madsen, M.M. Solomon, Recent developments in dynamic Operations Research 61 (1) (1995) 143–164.
vehicle routing systems, in: The Vehicle Routing Problem: Latest Advances [119] V. Pureza, G. Laporte, Waiting and buffering strategies for the dynamic pickup
and New Challenges, Operations Research/Computer Science Interfaces and delivery problem with time windows, INFOR 46 (3) (2008) 165–175.
Series, vol. 43, Springer, US, 2008, pp. 199–218. [120] A. Regan, H. Mahmassani, P. Jaillet, Improving efficiency of commercial
[92] J.-Q. Li, P.B. Mirchandani, D. Borenstein, A lagrangian heuristic for the real- vehicle operations using real-time information: potential uses and
time vehicle rescheduling problem, Transportation Research Part E: Logistics assignment strategies, Transportation Research Record: Journal of the
and Transportation Review 45 (3) (2009) 419–433. Transportation Research Board 1493 (1995) 188–198.
[93] J.-Q. Li, P.B. Mirchandani, D. Borenstein, Real-time vehicle rerouting problems [121] A. Regan, H. Mahmassani, P. Jaillet, Evaluation of dynamic fleet management
with time windows, European Journal of Operational Research 194 (3) (2009) systems – simulation framework, in: Forecasting, Travel Behavior, And
711–727. Network Modeling, number 1645 in Transportation Research Record, 1998,
[94] S. Lorini, J.-Y. Potvin, N. Zufferey, Online vehicle routing and scheduling with pp. 176–184.
dynamic travel times, Computers & Operations Research 38 (7) (2011) 1086– [122] A.C. Regan, H.S. Mahmassani, P. Jaillet, Dynamic decision making for
1090. commercial fleet operations using real-time information, Transportation
[95] K. Lund, O.B.G. Madsen, J.M. Rygaard, Vehicle routing problems with varying Research Record: Journal of the Transportation Research Board 1537 (1996)
degrees of dynamism, Technical report, IMM Institute of Mathematical 91–97.
Modelling, 1996. [123] A. Rizzoli, R. Montemanni, E. Lucibello, L. Gambardella, Ant colony
[96] E. Melachrinoudis, A.B. Ilhan, H. Min, A dial-a-ride problem for client optimization for real-world vehicle routing problems, Swarm Intelligence 1
transportation in a health-care organization, Computers & Operations (2007) 135–151.
Research 34 (3) (2007) 742–759. [124] M. Romero, L. Sheremetov, A. Soriano, A genetic algorithm for the pickup and
[97] J.E. Mendoza, B. Castanier, C. Guéret, A.L. Medaglia, N. Velasco, A memetic delivery problem: An application to the helicopter offshore transportation,
algorithm for the multi-compartment vehicle routing problem with in: Theoretical Advances and Applications of Fuzzy Logic and Soft Computing,
stochastic demands, Computers & Operations Research 37 (11) (2010) Advances in Soft Computing, vol. 42, Springer, Berlin/Heidelberg, 2007, pp.
1886–1898. 435–444.
[98] J.E. Mendoza, B. Castanier, C. Guéret, A.L. Medaglia, N. Velasco, Constructive [125] J. Roy, Recent trends in logistics and the need for real-time decision tools in
heuristics for the multicompartment vehicle routing problem with stochastic the trucking industry, in: Proceedings of the 34th Annual Hawaii
demands, Transportation Science 45 (3) (2011) 346–363. International Conference on System Sciences, 2001, 2001.
[99] M. Mes, M. Van der Heijden, A. Van Harten, Comparison of agent- [126] N. Secomandi, Comparing neuro-dynamic programming algorithms for the
based scheduling to look-ahead heuristics for real-time transportation vehicle routing problem with stochastic demands, Computers & Operations
problems, European Journal of Operational Research 181 (1) (2007) 59– Research 27 (11–12) (2000) 1201–1225.
75. [127] N. Secomandi, A rollout policy for the vehicle routing problem with stochastic
[100] S. Mitrović-Minić, R. Krishnamurti, G. Laporte, Double-horizon based demands, Operations Research 49 (5) (2001) 796–802.
heuristics for the dynamic pickup and delivery problem with time [128] N. Secomandi, F. Margot, Reoptimization approaches for the vehicle-routing
windows, Transportation Research Part B: Methodological 38 (8) (2004) problem with stochastic demands, Operations Research 57 (1) (2009) 214–
669–685. 230.
[101] S. Mitrović-Minić, G. Laporte, Waiting strategies for the dynamic pickup and [129] H. Simao, J. Day, A. George, T. Gifford, J. Nienow, W.B. Powell, An approximate
delivery problem with time windows, Transportation Research Part B: dynamic programming algorithm for large-scale fleet management: a case
Methodological 38 (7) (2004) 635–655. application, Transportation Science 43 (2) (2009) 178–197.
V. Pillac et al. / European Journal of Operational Research 225 (2013) 1–11 11
[130] A. Slater, Specification for a dynamic vehicle routing and scheduling system, Smith, J.J. Merelo-Guervós, J.A. Bullinaria, J. Rowe, P. Tino, A. Kabán, H.-P.
International Journal of Transport Management 1 (1) (2002) 29–40. Schwefel (Eds.), Parallel Problem Solving from Nature, Lecture Notes in
[131] D. Sleator, R. Tarjan, Amortized efficiency of list update and paging rules, Computer Science, vol. 3242, Springer, Berlin/Heidelberg, 2004, pp. 692–701.
Communications of the ACM 28 (2) (1985) 202–208. [145] P. Van Hentenryck, R. Bent, Online Stochastic Combinatorial Optimization,
[132] N. Smolic-Rocak, S. Bogdan, Z. Kovacic, T. Petrovic, Time windows based MIT Press, 2006.
dynamic routing in multi-agv systems, IEEE Transactions on Automation [146] B. Verweij, S. Ahmed, A. Kleywegt, G. Nemhauser, A. Shapiro, The sample
Science and Engineering 7 (1) (2010) 151–155. average approximation method applied to stochastic routing problems: a
[133] M.M. Solomon, Algorithms for the vehicle-routing and scheduling problems computational study, Computational Optimization and Applications 24 (2)
with time window constraints, Operations Research 35 (2) (1987) 254–265. (2003) 289–333.
[134] M. Spivey, W.B. Powell, The dynamic assignment problem, Transportation [147] C. Waters, Vehicle-scheduling problems with uncertainty and omitted
Science 38 (4) (2004) 399–419. customer, The Journal of the Operational Research Society 40 (12) (1989)
[135] R. Stahlbock, S. Voss, Operations research at container terminals: a literature 1099–1108.
update, OR Spectrum 30 (1) (2008) 1–52. [148] N. Wilson, N. Colvin, Computer control of the Rochester dial-a-ride system,
[136] M. Tagmouti, M. Gendreau, J.-Y. Potvin, A dynamic capacitated arc routing Technical Report Report R77-31, Dept. of Civil Engineering, Massachusetts
problem with time-dependent service costs, Transportation Research Part C: Institute of Technology, Cambridge, Massachusetts.
Emerging Technologies 19 (1) (2011) 20–28. [149] Z. Xiang, C. Chu, H. Chen, The study of a dynamic dial-a-ride problem under
[137] E. Taillard, P. Badeau, M. Gendreau, F. Guertin, J. Potvin, A tabu search time-dependent and stochastic environments, European Journal of
heuristic for the vehicle routing problem with soft time windows, Operational Research 185 (2) (2008) 534–551.
Transportation Science 31 (2) (1997) 170–186. [150] J. Yang, P. Jaillet, H. Mahmassani, Real-time multivehicle truckload pickup
[138] E.D. Taillard, L.M. Gambardella, M. Gendreau, J.-Y. Potvin, Adaptive memory and delivery problems, Transportation Science 38 (2) (2004) 135–148.
programming: a unified view of metaheuristics, European Journal of [151] S. Yang, M. Hamedi, A. Haghani, Online dispatching and routing model for
Operational Research 135 (1) (2001) 1–16. emergency vehicles with area coverage constraints, in: Network Modeling
[139] E. Taniguchi, H. Shimamoto, Intelligent transportation system based dynamic 2005, number 1923 in Transportation Research Record, 2005, pp. 1–8.
vehicle routing and scheduling with variable travel times, Transportation [152] Y. Yao, O. Ergun, E. Johnson, Integrated model for the dynamic on-demand air
Research Part C: Emerging Technologies 12 (3–4) (2004) 235–250. transportation operations, in: V. Zeimpekis, C.D. Tarantilis, G.M. Giaglis, I.
[140] E. Taniguchi, R. Thompson, Modeling city logistics, Transportation Research Minis (Eds.), Dynamic Fleet Management, Operations Research/Computer
Record: Journal of the Transportation Research Board 1790 (1) (2002) 5–51. Science Interfaces Series, vol. 38, Springer, US, 2007, pp. 95–111.
[141] B.W. Thomas, Waiting strategies for anticipating service requests from [153] V. Zeimpekis, I. Minis, K. Mamassis, G.M. Giaglis, Dynamic management of a
known customer locations, Transportation Science 41 (3) (2007) 319–331. delayed delivery vehicle in a city logistics environment, in: V. Zeimpekis, C.D.
[142] B.W. Thomas, Chelsea C.I. White, Anticipatory route selection, Transportation Tarantilis, G.M. Giaglis, I. Minis (Eds.), Dynamic Fleet Management,
Science 38 (4) (2004) 473–487. Operations Research/Computer Science Interfaces Series, vol. 38, Springer,
[143] P. Toth, D. Vigo (Eds.) The Vehicle Routing Problem, SIAM Monographs on US, 2007, pp. 197–217 (chapter 9).
Discrete Mathematics, vol. 9, SIAM Philadelphia, 2002. [154] V. Zeimpekis, C.D. Tarantilis, G.M. Giaglis, I. Minis (Eds.), Dynamic Fleet
[144] J.I. Van Hemert, J.L. Poutré, Dynamic routing problems with fruitful regions: Management, Operations Research Computer Science Interfaces Series, vol.
models and evolutionary computation, in: X. Yao, E. Burke, J.A. Lozano, J. 38, Springer, US, 2007.