2012 American Control Conference

Fairmont Queen Elizabeth, Montréal, Canada

June 27-June 29, 2012

Ant Colony Optimization Technique to Solve the Min-Max Multi

Depot Vehicle Routing Problem
Koushik S Venkata Narasimha, Elad Kivelevitch, and Manish Kumar, Member, ASME

Abstract— In this paper, we extend our work on solving min- Problem and involves finding tours of vehicles (with
max single depot vehicle routing, published in the proceedings constraints on maximum distance they can travel) from a
of the ACC 2011, to solving min-max multi depot vehicle depot that visits a given number of delivery points and
routing problem. The min-max multi-depot vehicle routing minimizes the total distance travelled. Solving this problem,
problem involves minimizing the maximum distance travelled
by any vehicle in case of vehicles starting from multiple depots
just like classical Travelling Salesman Problem (TSP), is
and travelling to each customer location (or city) at least once. computationally extensive and is known to be a
This problem is of specific significance in case of time critical Nondeterministic Polynomial (NP) hard problem. Multi-
applications such as emergency response in large-scale Depot Vehicle Routing Problem (MDVRP) extends the
disasters, and server-client network latency. In this paper we SDVRP by having multiple depots where multiple vehicles
extend the ant colony based algorithm which was proposed can originate from.
earlier in our previous paper and introduce a novel way to
address the min-max multi-depot vehicle routing problem. The
An interesting variant of this vehicle routing problem is a
approach uses a region partitioning method developed by min-max multi depot vehicle routing problem (min-max
Carlsson et al. to convert the multi-depot problem into multiple MDVRP). The objective of the problem is to minimize the
single-depot versions. A computer simulation model using maximum distance travelled by a vehicle instead of the total
MATLAB was developed. Also, in terms of optimality of distance travelled which is the case in the conventional
solution and computational time, a comparison with the existing MDVRP. This paper is an extension of the solution proposed
Carlsson model has been carried out. for min-max Single Depot Vehicle Routing Problem
(SDVRP) which was published in the proceedings of the
I. INTRODUCTION 2011 ACC at San Francisco [4]. This problem is often of

E fficient transportation of products in and around the

globe is an essential part of economy mainly because of
its high cost impact by reduction of energy consumption and
interest when minimization of time taken to visit all points is
more important than the total distance travelled, which is a
usual factor of interest in emergency management situations.
by increasing customer satisfaction. In the last decade, In emergency management, the objective is to use all
research suggests that 10% to 15% of the traded goods available vehicles to minimize the time taken to attend to all
correspond to the transportation costs. Also, the U.S. Bureau points needing emergency resources. Other applications of
of Labor Statistics estimates that transportation-related fields this problem are in defense and computer networking. For
example, assigning tours to a group of UAVs engaged in
are growing by nearly 56,000 jobs a year, thus showing an
large scale surveillance operation by solving min-max
increase in trade and logistic businesses [1]. Realizing the
problem will minimize the maximum time of travel of UAVs,
importance of this factor, researchers have devoted a lot of
and hence help achieve desired objectives in time-critical
effort in finding out novel and optimal ways for an efficient scenarios. In computer networking, depots represent servers,
transportation. According to Toth and Vigo in [2], utilization vehicles represent data packets and customers represent
of computerized methods for transportation will result in clients. A network routing topology generated by solving the
significant savings ranging from 5% - 20%. A well known min-max problem would result in minimizing the maximum
problem in this field that has emerged as a benchmark latency between any pair of server and a client. In the next
optimization problem during the past few decades is the section, a brief background on the previous researches
Single Depot Vehicle Routing Problem (SDVRP). This available in the literature has been provided. Also discussed
problem was first proposed byDantzig and Ramster in 1959 is our previous research on min-max SDVRP.
[3]. It is an extension of the classical Travelling Salesman
II. BACKGROUND
Manuscript received September 23, 2011.
Koushik S Venkata Narasimha is pursuing his graduate studies at the
A. Our previous paper on min-max SDVRP
University of Cincinnati, OH 45221 USA (phone: 309-339-3245; e-mail: The previous paper [4] was focused on solving the min-
venkatkh@mail.uc.edu). max Single Depot Vehicle Routing Problem in which the
Elad Kivelevitch is with the University of Cincinnati, OH 45221 USA.
He is a PhD student at School of Aerospace Systems (e-mail: objective was to minimize the maximum length of a tour of a
kiveleeh@ucmail.uc.edu). traditional SDVRP. Ant colony optimization technique was
Manish Kumar is with the University of Cincinnati, OH 45221. He is an used to solve this problem which was based on the approach
Assistant Professor at the School of Dynamic Systems. (e-mail:
kumarmu@ucmail.uc.edu). used by Bullnheimer et al. [5] to solve the traditional

978-1-4577-1096-4/12/$26.00 ©2012 AACC 3980

SDVRP. In this approach, the artificial or simulated ants MDVRP with uniformly distributed points would
construct vehicle routes by successively choosing the next numerically approach a value proportional to n / k , which
cities using a probabilistic transition rule. If the probability was the value of optimal TSP tour of all customers split by
‘p’ value for transition to a particular jth city from the ith city number of vehicles ‘k’, under the constraint. Later they
is higher, then chances of that city getting chosen as the next developed two different heuristics to solve the min-max
city is increased. During every transition, the tour length is MDVRP. The first heuristic was a linear programming based
calculated and whenever the choice of next city would lead load balancing technique which is of less relevance to our
to an infeasible solution, i.e., if the tour length exceeds the proposed paper. The second and an intriguing heuristic is the
vehicle distance constraint (the maximum distance that a region partition based method. In this technique, noticing
vehicle can travel) L, then the depot is chosen as the next city that a convex equitable partition yields an even division of
(so that the tour of that vehicle is finished) and a new tour is points, they divided the service region into a set of
started for a new vehicle. The vehicle capacity L is a critical subregions with equal area and generated good initial
parameter in finding out the min-max solution to the solutions by assigning the customer points in the depot
proposed problem. This is explained further in the Approach region to the respective depot.
section of this paper.
D. Region Partitioning
B. Multi Depot Vehicle Routing Problem (MDVRP)
A brief overview of the studies that have been carried out
MDVRP can be traced back to 1976 when Gillet and in finding equitable partitions of a region is presented in this
Johnson published a paper on Multi Terminal Vehicle- section. This is pertinent as our solution implements these
Dispatch Algorithm [6]. In this paper, a heuristic algorithm methods in generating a solution to the min-max MDVRP. In
was developed to obtain an approximate solution. Their 1942, Stone and Tukey [12] proposed a basic partition
purpose was to determine a set of routes by which vehicles theorem called Ham Sandwich Theorem (HST) which states
from two or more terminals could service a collection of that “given n measurable objects in n-dimensional space, it is
demand points while minimizing the total distance travelled. possible to divide all of them in half with respect to their
A sweep algorithm was employed based on the strategy to measure with a (n-1)-dimensional hyperplane”. In a
partition the problem to single-terminal problem in order to computational geometry perspective as proposed by [13], a
significantly reduce the computation time. The solution was discrete HST can be interpreted as follows. For a planar case
also extended to satisfy some of the constraints such as the of finite sets of red and blue points, there exists a line
vehicle load and the length of each route. After this paper, dividing both red and blue points into sets of equal size.
much effort has been dedicated by researchers around the A generalized Ham Sandwich Theorem was proven by
globe and many have come up with different methods to Bespamyatnikh et al. in [14]. In this paper, they
solve this problem in an optimal way. [7], [8], [9] are some demonstrated that given gn red points and gm blue points in
of the different methods that were used to solve MDVRP. In the plane in general position, there existed an equitable
2005, Lim and Wang [10] proposed a more practical variant subdivision of the plane into g disjoint convex polygons each
of this problem and it was named MDVRP with fixed of which contains n red points and m blue points. The Ham
distribution of vehicles (MDVRPFD). They proposed this Sandwich Theorem is a subset of this problem given the
problem with a bound on the number of vehicles in a depot specific case of g=2.
unlike the traditional MDVRP where the limit was On the basis of the generalized Ham Sandwich Theorem,
unrealizable infinite number of vehicles. With an assumption Carlsson et al. developed an algorithm to find equitable
of exactly one vehicle in each depot, they developed a binary convex partitions in a polygon [15]. This problem was
programming model to obtain the solution and generalized addressed with a motivation to obtain a solution to the
the solution for n number of vehicles in a depot. In this MDVRP. Region partition is a key step in this vehicle
paper, they proposed a new one-stage approach where in the routing problem as the customers need to be allocated to
assignment of customers to depots and route calculations vehicles in a load balancing manner. For a min-max type of a
were done at the same level. problem, since the travel times of the vehicle has to be
C. Min-Max MDVRP minimum in order to obtain the optimal solution, partitioning
the polygon equally is the primal factor. This method is
An important and interesting variant of VRP, closely
further explained in the Approach section.
related to VRP with time windows (VRPTW), is called Min-
Max MDVRP. This problem was first formulated by
III. PROBLEM DESCRIPTION
Carlsson et al. in 2007 [11]. Unlike the traditional objective
of minimizing the tour lengths of the vehicles, the purpose of The paper focuses on the min-max MDVRP in which the
this problem is to minimize the maximum length of a tour in objective is to minimize the maximal length of a tour in a
VRP. In this paper, they performed a theoretical analysis by traditional MDVRP. To mathematically formulate this
developing an asymptotic bound for longest tour length L problem, consider ‘n’ customer/delivery points, ‘m’ depot
and concluded that the optimal solution to the min-max points and a total number of ‘k’ vehicles at the depots. All

3981
the vehicles are initially located at their respective depots. the minimum L, called L*, that can still solve the traditional
The vehicles are required to visit all customer points and SDVRP using the existing vehicles. Eq. (5) below [11]
return to the same depot from where they started their represents a bound on L, and our algorithm attempts to find
journey. The problem as stated earlier is to decide the tours out the critical L* which can satisfy SDVRP condition.
of each vehicle so that the maximum distance travelled by Lemma 1: For a general planar graph representing the
any vehicle is minimized. Mathematically [11], the aim is to SDVRP,
Minimize 
(1) TSP( D  N )  TSP( D) TSP( N )
subject to TSP( S )   , i
i L  2  d ( D, N ) (5)
k k
 Si  N In Eq. 5, N denotes the set of customer points, D denotes
where N is the set of all customer locations, |N| = n, Si  N set of depots, k denotes number of vehicles in a depot,
is the subset of customers assigned to vehicle ‘i’ and TSP(Si) d(A,B) denotes the largest distance between an arbitrary pair
is the length of the optimal tour for a single Traveling of points in two different sets A and B i.e.,
Salesman Problem applied to vehicle i and the customer set d ( A, B)  max | | x  y | | . A proof of the above
Si. x  A, y  B
lemma is provided in [11].
IV. APPROACH With the min-max SDVRP solution established, the next
A detailed explanation on the ant colony approach that was step for extending the solution to multiple depots is to
used to get the solution to min-max SDVRP can be found in partition the region into equitable convex partitions. Once
Section IV of our previous paper [4], and is not presented in equitable partitions are found, solving the min-max MDVRP
this paper for brevity and adhering page limitations. We reduces to solving the min-max SDVRP for each partitioned
briefly highlight some of the important aspects of the region. However, it is necessary to achieve the partitioning in
algorithm relevant to the Ant Colony based algorithms [l6]. a way such that solving the min-max SDVRP for partitions
First is the heuristic desirability (nij). The heuristic would correspond to solving the complete min-max
desirability of visiting city j from city i (or the visibility) is MDVRP. In order to show that, we first present a result
represented by ηij and is equal to reciprocal of dij, the proven by Beardwood et al. [17] as a lemma below which
distance from city i to city j: can be used for arguing that, under certain assumptions,
1 partitioning the polygon (convex polygon hull of the entire
ij  (2)
d ij domain of customer points) equally would lead to an optimal
Second, the probabilistic transition rule (pij), which partitioning for our problem.
represents the probability of choosing city j from city i, and Lemma 2: If Xi ,1≤ i ≤ ∞ are independently and identically
is given by Eq.3. Here,ij is the pheromone concentration on distributed (i.i.d) random variables with bounded support in
path from i to j,Ω = vj  V is set of feasible cities, i.e., cities Rd, then the length Ln under the usual Euclidean metric of the
that have not been visited, α and β represent the biases for shortest path through the points {X1, X2,….Xn} satisfies
( d 1)
pheromone trail and visibility respectively. The use of pij Ln
provides a probabilistic mechanism to direct the search ( d 1)   TSPd  f ( x) d
dx (6)
based on combination of greedy, local choice (provided by n d R d

heuristic desirability) and global information (provided by Here f(x) is the probability density function representing the
pheromone concentration). distribution of points Xi, and βTSP,d is a positive constant that

 [ ij ] [ ij ]  depends on d but not on the distribution of the Xi.
pij   if v j  
  h [ ih ] [ ih ]

  Based on the above lemma, we can deduce a proof for our
(3) 2-dimensional problem [15]. Assuming the cities are
else pij  0 uniformly distributed and the function Σf(x)dx is the area of
Third, the pheromone update rule which is given by Eq. 4 the partitioned subregion A, Ln for our problem is the route
m length that was calculated and it is apparent with a few
 ijnew   ijold   
k 1
k *
ij   ij (4) substitutions (d=2, n = ‘k’ no. of cities), that route length of
each vehicle is asymptotically proportional to the square root
The simulated ants construct the solution (vehicle routes) of the area of subregion. Mathematically,
1
based on the above three functions and during every Ln (7)
transition, the route length (L) is calculated and when this 1   TSPd A 2
exceeds the vehicle distance constraint (or the capacity Lmax), n 2

then the depot is chosen as the next city and the route Thus establishing an asymptotic relation between tour length
calculation will start fresh again. Thus this constraint on L and area, it is optimal to divide the area into equitable
acts as a critical parameter in finding out the min-max partitions as the lengths of optimal tours in each partition
solution. Solving the min-max SDVRP boils down to finding would approximately be the same. Because, for example, if

3982
an area of subregion (A1) is larger than subregion (A2), then TABLE I
route length (L1) of that area A1 becomes longer and hence ANT COLONY OPTIMIZATION ALGORITHM TO SOLVE MIN-MAX MDVRP
the time taken to cater to that area increases which should
not be the case since it is a min-max problem. Hence the /*Inputs*/
most optimal way is to divide the area into equal partitions. 1. cust_points = get_custpoints() /*set of x,y coordinates of the
cities */
Going back to our step of partitioning the given area, based
2. depot = get_depotpoints()/*set of x,y coordinates of the
on the above theorem we have the following important depots*/
assumption. Customer points in the 2-d plane are uniformly /*Generate convex hull*/
distributed. A well known partitioning method in the field of 3. poly_points = [cust_points; depot]/*set of combination of
depot and cust_points*/
computational geometry is the Voronoi diagram. The method 4. poly_vertices = convhull(poly_points) /*get the vertices of
in particular is the Centroidal Voronoi Tessellation. In this the convex hull using convhull function*/
method, first, given points are separated based on the relative /*Region Partitioning using Carlsson algorithm*/
5. subregions = region_partition(poly_vertices,depot)/*get the
closeness/density. Second, a centroidal point is calculated. vertices of the partitioned polygon and also get the depot
Finally, a hyperplane is drawn by considering the fact that points corresponding to these set of subregion vertices*/
the set of points that are closer to centroid belong to one /* Main Loop*/
no_of_depots = size(depot)
voronoi cell. In our case these cells can be considered as
6. For i = 1 to no_of_depots do
subregions. But the drawback of this method is that it does IN = inpolygon()/*assign the cust_points that are in
not always divide the area into equal partitions. And if the partition i to depot i, basically check the points if it inside or
partitions are not of same area, then we have already proved outside the polygon which is in focus*/
V = cust_points(IN)/*V is the customer points that are
that it will be a suboptimal solution. Hence we implement a inside region ‘i’*/
different method that was proposed by Carlsson et al. This V0 = depot(i)/*depot(i) corresponds to the x,y coordinate
method uses an approximation algorithm to find the location of the ith depot
/*Generate the min-max SDVRP tour and calculate
of partition by performing binary searches over the given set optimum tour length and optimum vehicle tour for each
of depot points and customer points. A complete explanation vehicle*/
of this method is out of scope of this paper and can be found [opt_vehic_tour opt_tour_length] = sdvrp(V,V0)
End For
in Sections 2 and 3 of [11]. Based on this method any given
/*Plot the Vehicle tours*/
polygon can be partitioned so that: 7. For i = 1 to no_of_depots do
a) All the partitions are convex polygons. For k = 1 to no_of_vehic do
b) All the partitions contain exactly one depot plot()/*plot the vehicle tour*/
End For
c) All the partitions have equal area End For
A sample figure representing a partitioned region that was
obtained using the above method is shown in Fig. 1. V. RESULTS AND DISCUSSIONS
Explained in Table I is the proposed algorithm to obtain an
Before we discuss the results, let us capture some of the
approximate solution to min-max MDVRP. It may be
assumptions that are made: a) Customer points and depot
mentioned that text between /* and */ represent the
points are uniformly distributed over 2-d space b) the vehicle
comments explaining the steps. Also, the subroutine
capacities remain the same throughout all depots.
sdvrp(V,V0) is the subroutine representing the min-max
SDVRP solution and can be found in [4].

(A) (B)
Fig 2: For the case of 80 Customers and 4 depots (A) Results
obtained using Carlsson based approach (B) Results obtained
using Ant Colony based approach
A MATLAB based computer program has been developed
in order to validate our proposed solution and a comparison
with Carlsson based method [11, 18] has been carried out to
check the credibility. Fig. 2 compares the solution for 4
depot-80 city problem between Carlsson based method and
Fig 1: A convex equitable partition for a case of 10 depots. our Ant Colony based method. Each depot has two vehicles.
Table II compares the distances traversed by each vehicle for

3983
the 4 depot – 80 city problem. The maximum vehicle In practice, the cities may not be distributed uniformly, in
distance in ant colony based method is by vehicle 1 in depot which case the results may not be as optimal. However, sub-
3 which is 125.35 units. Comparing with the Carlsson’s optimal solutions can be used with other optimization
method where the maximum distance is 131.47 units (vehicle methods to obtain a near-optimal solution. Moreover, this
2 in 2nd depot), it is clear that ant colony based method approximation will get better as the total number of cities
improves on the Carlsson’s method in this case. Two more
TABLE III
cases demonstrating improved performance of the ant colony
solution over the Carlsson algorithm in terms of optimality of COMPARSION OF ANT COLONY BASED AND CARLSSON BASED TECHNIQUE TO
SOLVE MIN-MAX MDVRP FOR A 4 DEPOT 140 CITY PROBLEM
TABLE II
Distance travelled by
each Distance travelled by
COMPARSION OF ANT COLONY BASED AND CARLSSON BASED TECHNIQUE TO Depot
vehicle(L_opt_vehic) each vehicle (Carlsson
SOLVE MIN-MAX MDVRP FOR A 4 DEPOT 80 CITY PROBLEM No.
(Ant Colony based based method)
Distance travelled by method)
each Distance travelled by 1 Vehic 1 = 110.97 Vehic 1 = 97.91
Depot
vehicle(L_opt_vehic) each vehicle (Carlsson Vehic 2 = 114 Vehic 2 = 111.65
No.
(Ant Colony based based method) Vehic 3 = 25.24 Vehic 3 = 114.39
method)
1 Vehic 1 = 111.60 Vehic 1 = 114.05 2 Vehic 1 = 114.57 Vehic 1 = 171.40*
Vehic 2 = 129.20 Vehic 2 = 110.96 Vehic 2 = 103.83 Vehic 2 = 105.09
Vehic 3 = 95.11 Vehic 3 = 85.85
2 Vehic 1 = 94.21 Vehic 1 = 108.48
Vehic 2 = 100.39 Vehic 2 = 131.47* 3 Vehic 1 = 115.94 Vehic 1 = 120.27
Vehic 2 = 105.66 Vehic 2 = 142.12
3 Vehic 1 = 125.35* Vehic 1 = 104.38 Vehic 3 = 102.22 Vehic 3 = 113.29
Vehic 2 = 114.59 Vehic 2 = 92.37
4 Vehic 1 = 116.28* Vehic 1 = 82.57
4 Vehic 1 = 118.97 Vehic 1 = 102 Vehic 2 = 112.56 Vehic 2 = 79.89
Vehic 2 = 97.71 Vehic 2 = 124.65 Vehic 3 = 91.21 Vehic 3 = 67.70

* denotes the maximum distance travelled in a set of vehicle routes. * denotes the maximum distance travelled in a set of vehicle routes.

solution are shown with the results tabulated in Table III and increases Thus, this is a good initial guess for the solution,
Table IV. Fig 3 and Fig 4 represent these cases respectively. but for smaller sets of cities there will probably be a need to
Table III (and Fig. 3) represents 4 depot-140 city (each exchange cities between tours after this initial partitioning is
depot has 3 vehicles) problem, and Table IV (Fig. 4) done. A drawback of this method is the calculation time
represents 5 depot-140 city problem. Also, it can be required to achieve the solution. For the cases from Fig 2 –
observed that the vehicle lengths in the ant colony approach Fig 4, the approximate time taken to achieve the solution is
are almost uniform (few exceptions) validating the claim 200 – 250 seconds. On the other hand, Carlsson method was
made using Lemma 2 above that equal partitioning would able to solve in few seconds.
lead to equitable distributions of cities and hence the lengths
of the tours of cities. Since the area has been partitioned
equally, these lengths are also distributed equally. Getting a
guaranteed optimal solution to this problem is known to be
NP hard, however approximate methods, such as proposed
here, can significantly improve the time required to calculate

(A) (B)
Fig 4: For the case of 140 customers and 5 depots. (A)
Results obtained using Carlsson based approach (B) Results
obtained using Ant Colony based approach

This is because the method used by Carlsson et al. utilizes

(A) (B)
the well-known Concorde TSP solver [19] which can get to
Fig 3: For the case of 140 customers and 4 depots. (A)
near-exact TSP solution within a few seconds. One way to
Results obtained using Carlsson based approach (B) Results
improve the running time is to use the inherent distributed
obtained using Ant Colony based approach
nature of the problem once the area is partitioned, and to
a solution. It may be noted that the paper makes an perform the calculation of the each SDVRP on a separate
assumption regarding the uniform distribution of cities. processor.

3984
 Problem”, American Control Conference(ACC), pp. 3257-3262,
TABLE IV 2011.
[5] Bullnheimer, B., Hartl, R.F., Strauss, C., “Applying the Ant System to
COMPARSION OF ANT COLONY BASED AND CARLSSON BASED TECHNIQUE TO the Vehicle Routing Problem”, Proceedings of 2nd International
SOLVE MIN-MAX MDVRP FOR A 5 DEPOT 140 CITY PROBLEM Conference on Metaheuristics, 1997.
[6] Gillett, B.E., Johnson, J.G., “Multi-Terminal Vehicle-Dispatch
Distance travelled by
Algorithm”, Omega, vol.4, pp. 771-718, 1976.
each Distance travelled by
Depot [7] Tillman, F., “The multiple terminal delivery problem with
vehicle(L_opt_vehic) each vehicle (Carlsson
No. probabilistic demands”, Transportation Science, vol. 3, pp. 192-204,
(Ant Colony based based method)
1969.
method)
[8] Chao, I.M., Golden, B., Wasil, E., “A new heuristic for the multidepot
1 Vehic 1 = 108.15 Vehic 1 = 89
vehicle routing problem that improves upon best-known solutions”,
Vehic 2 = 96.66 Vehic 2 = 94.48
American Journal of Mathematical and Management Sciences, vol.
13, no. 3, pp. 371-406, 1993.
2 Vehic 1 = 117.17 Vehic 1 = 89.41
[9] Laporte, G., Nobert, Y., Taillefer, S., “Solving a family of multi-depot
Vehic 2 = 104.08 Vehic 2 = 96.95
vehicle routing and location-routing problems”, Transportation
Science, vol. 22, p.172, 1988.
3 Vehic 1 = 114.02 Vehic 1 = 73.46
[10] Lim, A., Wang., “Multiple-Depot Vehicle Routing Problem: A One-
Vehic 2 = 116.20 Vehic 2 = 94.47
Stage Approach”, IEEE Transactions on Automation Science and
Vehic 3 = 46.99 Vehic 3 = 110.20
Engineering, vol. 2, no. 4, pp. 397- 402, 2005.
[11] Carlsson, J., Ge, D., Subramaniam, A., Wu, A., Ye, Y., “Solving Min-
4 Vehic 1 = 118.03 Vehic 1 = 85.84
Max Multi-Depot Vehicle Routing Problem”, Fields Book on Global
Vehic 2 = 76.18 Vehic 2 = 126.05*
Optimization, 2007.
Vehic 3 = 113.65 Vehic 3 = 106.80
[12] Stone, A.H., Tukey, J.W., “Generalized sandwich theorems”, Duke
Mathematical Journal, vol. 9, pp. 356-359, 1942.
5 Vehic 1 = 81.90 Vehic 1 = 99.15
[13] Zivaljevic, R.T., “Topological Methods”, Handbook of Discrete and
Vehic 2 = 112.65 Vehic 2 = 104.73
Computational Geometry, chapter 11, pages 209-224, 1997.
Vehic 3 = 118.96* Vehic 3 = 65.25
[14] Bespamyatnikh, S., Kirkpatrick, D.G., Snoeyink, J., “Generalizing
Ham Sandwich Cuts to Equitable Subdivisions”, Symposium on
* denotes the maximum distance travelled in a set of vehicle routes. Computational Geometry (SoCG), pp. 49-58, 1999.
[15] Carlsson, J., Armbruster, B., Ye, Y., “Finding equitable convex
VI. CONCLUSIONS AND FUTURE SCOPE partitions of points in a polygon efficiently”, ACM Transactions on
Algorithms, vol. 6, issue 4, article no. 72, 2010.
The paper presents an ant colony optimization based [16] Dorigo, M., Gambardella, L.M., “Ant Colonies for the Travelling
approximate algorithm to solve the min-max MDVRP by Salesman Problem”, BioSystems 43, pp. 73-81, 1997.
[17] Beardwood, J., Halton, J.H., Hammersley, J.M., “The shortest path
extending our previous solution from single depot to through many points”, Proceedings of Cambridge Philosophical
multiple depots. Unlike the traditional MDVRP, which Society, vol. 55, pp. 299-327, 1959.
minimizes the total distance travelled, the min-max MDVRP [18] http://www.stanford.edu/class/msande310/handout.shtml
[19] http://www.tsp.gatech.edu/concorde.html
minimizes the maximal distance travelled by a vehicle. This
version of problem holds immense applications for time-
critical problems. The problem was subdivided into many
min-max SDVRPs by partitioning equally the region
consisting of depots and cities. The proposed technique has
been validated with theory and with extensive simulations,
and results have been compared with the results obtained
using a standard method given by Carlsson. The results
obtained using the proposed ant colony based approach
provide improvements in terms of optimality over the
standard approach. Future work includes using parallel
processing programming techniques to expedite the
convergence towards the solution and integrating this
method with other optimization techniques to achieve more
optimal solutions in cases where we have non-uniform
distribution or smaller number of cities.

REFERENCES
[1] US Department of Labor, “Occupational Outlook Handbook”, Bureau
of Labor Statistics, 2010-11.
[2] Toth,P., Vigo, D., "The Vehicle Routing Problem", Monographs on
Discrete Mathematics and Applications, SIAM, Philadelphia, 2001.
[3] Dantzig, G.B., Ramser J.H., “The truck dispatching problem”,
Management Science, Vol. 6:60, 1959.
[4] Venkata Narasimha, K.S., Kumar, M., “Ant Colony Optimization
Technique to Solve the Min-Max Single Depot Vehicle Routing

3985