Scribd
Upload
14 views5 pages

The Last Mile Delivery Problem

Uploaded by

gokudeptrai1
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
14 views5 pages

The Last Mile Delivery Problem

Uploaded by

gokudeptrai1
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

The last mile delivery problem

Yuan Yuan, Diego Cattaruzza, Maxime Ogier, Frédéric Semet

To cite this version:


Yuan Yuan, Diego Cattaruzza, Maxime Ogier, Frédéric Semet. The last mile delivery problem.
ROUTE 2018 - International Workshop on Vehicle Routing, Intermodal Transportation and Related
Areas, May 2018, Snekkersten, Denmark. ฀hal-01964234฀

HAL Id: hal-01964234


https://hal.inria.fr/hal-01964234
Submitted on 21 Dec 2018

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est


archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
The last mile delivery problem
Yuan Yuan, Diego Cattaruzza, Maxime Ogier, Frédéric Semet
Univ. Lille, CNRS, Centrale Lille, Inria,
UMR 9189 - CRIStAL Lille, France
Email: [email protected]

1 Introduction and problem definition


E-commerce is a thriving market around the world and suits very well the busy lifestyle of
today’s customers. An annual survey by analytics firm comScore and UPS revealed that
consumers in US were purchasing more things online than in stores in 20161 . According
to the Ecommerce fondation, 1.4 billion people purchased goods and/or services online
at least once in 2015. They spent $ 2,272.7 billion online, which results in an average
spending per e-shopper of $ 1,582. It is obvious that the e-commerce growth poses a huge
challenge for transportation companies, especially in the last mile delivery. According
to [1], the last mile parcel delivery cost often reaches or even exceeds 50% of the total
transportation cost, making it a top concern for many companies.
Nowadays, the most common last mile delivery service is home delivery. Customers
wait at home to get their orders. Besides home delivery, companies like Amazon and
Fedex, develop locker and pick-up&go delivery services. When customers shop online,
they can choose a nearby locker or a store offering a pick-up&go counter. In the past two
years, a new concept called trunk delivery, has been proposed. Here, customers’ orders
can be delivered to the trunk of their cars. Volvo launched its in-car delivery service in
Sweden in 2016. The courier has a one-time digital code to get access to the car.
Trunk delivery is different from home delivery and locker delivery since the car moves
during the day and can be in different locations during the planning horizon. We study an
efficient last mile delivery system that combines all these delivery services: home, locker,
pick-up&go location and car trunk.
In this presentation, we address the routing problem for one vehicle. The problem
is modeled on a directed graph G = (V, A). The set of vertices V = {0, 1, ..., N } is
partitioned into C0 = {0}, C1 , ..., CK clusters. Cluster C0 contains only the depot. Each
other cluster Ck , k > 0 represents the set of alternative locations on which client k can
1
http://fortune.com/2016/06/08/online-shopping-increases/
be delivered. Each vertex is associated with a time-window [Ei , Li ], i ∈ {0, 1, ..., N } with
[E0 , L0 ] = [0, T ]. A visit can only be made to a vertex during its time-window and an
early arrival leads to waiting time while a late arrival causes infeasibility. In a cluster, the
time-windows associated with the vertices may overlap. Arcs are only defined between
vertices belonging to different clusters, that is, A = {(i, j) : i ∈ Ck , j ∈ Cl , k 6= l}. Each
arc (i, j) ∈ A is associated with a traveling cost Cij and time Tij . A tour starts from and
ends at the depot.
The objective is to find a minimum cost tour visiting each customer at one location
within the associated time-window. The problem that arises is called the Generalized Trav-
eling Salesman Problem with Time Windows (GTSPTW). We assume that this problem is
static and deterministic, namely all customer locations and the associated time-windows
are known with certainty in advance. The GTSPTW can be modeled as follows.

X
min Cij xij (1)
(i,j)∈A
X
s.t. yi = 1 k ∈ {0, 1, ..., K},
i∈Ck
(2)
X X
xij = xji = yi ∀i ∈ V,
(i,j)∈δ + (i) (j,i)∈δ − (i)
(3)
X X
E i y i ≤ tk ≤ L i yi k ∈ {0, 1, ..., K},
i∈Ck i∈Ck
(4)
X X
tk − tℓ + Tij xij ≤ L u yu − Ev yv − (Li − Ej )xij ∀(i, j) ∈ A, i ∈ Ck , j ∈ Cℓ ,
u∈Ck v∈Cℓ
(5)
yi ∈ {0, 1} ∀i ∈ V,
(6)
xij ∈ {0, 1} ∀(i, j) ∈ A.
(7)

where δ + (i) = {(i, j) ∈ A, j 6= i} and δ − (i) = {(j, i) ∈ A, j 6= i}. The objective func-
tion (1) minimizes the overall costs. Constraints (2) ensure that exactly one vertex from
each cluster is visited. Constraints (3) are flow conservation constraints. Constraints (4)
ensure that a vertex is visited during its time-window. Constraints (5) ensure that the ar-
rival and traveling times are consistent, meanwhile eliminating subtours. Constraints (6)
and (7) are variable definitions.
When time-windows are not considered, the GTSPTW reduces to the well-known
Generalized Travelling Salesman Problem (Fischetti et al. [2]). For the multi-vehicle case,
the problem is named Generalized Vehicle Routing Problem with TW (Moccia et al. [4]).
The special case where time-windows on clients do not overlap has been recently considered
by Reyes et al. [3]. The problem is called the Vehicle Routing Problem (VRP) with
Roaming Deliveries and models the case when only trunk deliveries are considered.

2 A branch-and-cut scheme
We propose several valid inequalities for GTSPTW. Some of them are derived from valid
inequalities for the asymmetric traveling salesman problem with time-windows, others are
specific. The main constraints are the followings :
P
• Feasible path inequality. xij + h∈Sij yh ≥ yi + yj − 1 i ∈ Ck , j ∈ C ℓ ,
where Sij = {h ∈ V \ (Ck ∪ Cℓ )|Ei + Tih ≤ Lh , Eh + Thj ≤ Lj , Ei + Tih + Thj ≤ Lj }
for i ∈ Ck and j ∈ Cℓ .
Two vertices are visited either directly or there exists a connection between them.

• Infeasible path elimination inequality. Several infeasible path elimination constraints


can be defined. The simplest ones are the followings:
P
xij + h∈C ij yh ≤ 1, k ∈ {1, . . . , K}, i, j ∈ V \ Ck , i 6= j
k
where Ckij = {h ∈ Ck |Eh +SPhi +Tij > Lj or Eh +SPhi > Li or Ei +Tij +SPjh > Lh
or Ej + SPjh > Lh , i, j ∈ V \ Ck }.
SPij represents the shortest traveling time from vertex i to vertex j. When the
triangle inequality is not satisfied, the shortest path from i to j can include the
visit of other vertices. These constraints detect if a vertex and an arc cannot be
simultaneously selected in a feasible solution due to time-windows.

• Generalized subtour elimination inequalities (GSECs).


P
• Clique inequality. i∈S yi ≤ |S| − 1.
If no feasible path passing through all the vertices of a set S ⊂ V exists (due to time-
windows), then the number of vertices of S that can be visited in all the feasible
solutions are less than the size of S.

We develop a branch-and-cut algorithm for the GTSPTW. We include at the root node
of the branch-and-bound tree all polynomial sets of inequalities while GSEC inequalities
and clique inequalities are separated in the course of the algorithm. An initial solution is
provided to the algorithm thanks to a heuristic which combines the generation of sequences
of clusters and the solution of shortest path problems with resources constraints.
The algorithm is implemented in C++ using Cplex 12.6 and the Concert technology.
Preliminary results are obtained on instances generated from GTSP instances. When
creating an instance, we guarantee that a feasible solution exists. We set the CPU time
limit to 1 hour. Instances with up to 22 clusters and 107 vertices are solved solved to
optimality. The average solution time for instances with less than 20 clusters is 19.4
seconds while it is equal to 441.6 seconds for instances with 20 and 22 clusters.

References
[1] M. Joerss, J. Schrder, F. Neuhaus, C. Klink and F. Mann, “Parcel Delivery the Future
of Last Mile”, Technical Report, McKinsey& Company, 2016.

[2] M. Fischetti, J.J. Salazar González and P. Toth, “A branch-and-cut algorithm for
the symmetric generalized traveling salesman problem”, Operations Research 45(3),
378–394, 1997.

[3] D. Reyes, M. Savelsbergh and A. Toriello, “Vehicle routing with roaming delivery
locations”, Transportation Research Part C: Emerging Technologies 80, 71–91, 2017.

[4] L. Moccia, J.F. Cordeau and G. Laporte, “An incremental tabu search heuristic for the
generalized vehicle routing problem with time windows”, Journal of the Operational
Research Society 63, 232–244, 2012.

You might also like