This article was downloaded by: [108.50.179.168] On: 19 April 2016, At: 18:23Publisher: Institute for Operations Research and the Management Sciences (INFORMS)INFORMS is located in Maryland, USA

Interfaces

Publication details, including instructions for authors and subscription information:http://pubsonline.informs.org

Transportation Planning and Scheduling for the 2014Special Olympics USA GamesAndrew Johnson, Yao Zhao, Xin Xu

To cite this article:Andrew Johnson, Yao Zhao, Xin Xu (2016) Transportation Planning and Scheduling for the 2014 Special Olympics USA Games.Interfaces

Published online in Articles in Advance 19 Apr 2016

. http://dx.doi.org/10.1287/inte.2016.0845

Full terms and conditions of use: http://pubsonline.informs.org/page/terms-and-conditions

This article may be used only for the purposes of research, teaching, and/or private study. Commercial useor systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisherapproval, unless otherwise noted. For more information, contact permissions@informs.org.

The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitnessfor a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, orinclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, orsupport of claims made of that product, publication, or service.

Copyright © 2016, INFORMS

Please scroll down for article—it is on subsequent pages

INFORMS is the largest professional society in the world for professionals in the fields of operations research, managementscience, and analytics.For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org

Articles in Advance, pp. 1–13ISSN 0092-2102 (print) � ISSN 1526-551X (online) http://dx.doi.org/10.1287/inte.2016.0845

© 2016 INFORMS

Transportation Planning and Scheduling for the2014 Special Olympics USA Games

Andrew JohnsonDepartment of Marketing and Business Information Systems, Rohrer College of Business, Rowan University,

Glassboro, New Jersey 08028, johnsonae@rowan.edu

Yao ZhaoDepartment of Supply Chain Management, Rutgers University, Newark, New Jersey 07102, yaozhao@andromeda.rutgers.edu

Xin XuFinancial Services, Ernst & Young, New York, New York 10036, xin.xu.2009@gmail.com

The 2014 Special Olympics USA Games were hosted in New Jersey. More than 4,000 athletes competed in 16 sportshosted across 10 locations within a 30-mile radius. We designed timely, convenient, easy-to-follow, and reliable busroutes and schedules to assist thousands of people with intellectual disabilities and their coaches to attend gamesand special events over seven days under a budget of $600,000.

Keywords : transportation planning; crew scheduling; Special Olympics; mega event; genetic algorithms; vehiclerouting and scheduling.

History : This paper was refereed. Published online in Articles in Advance.

Special Olympics is a nonprofit organization thatgrew out of Eunice Kennedy Shriver’s observations

of the unjust and unfair treatment of people withintellectual disabilities during the 1960s. The mission ofSpecial Olympics is to provide access to sports trainingand athletic competition for children and adults withintellectual disabilities so they can achieve, succeed,develop physical fitness, experience joy, and be anactive part of their local communities (Special Olympics,Inc. 2015). The global inspiration of all Special Olympicathletes is characterized by their powerful oath: “Letme win. But if I cannot win, let me be brave in theattempt” (Special Olympics: Our Athletes 2014).

The Special Olympics 2014 USA Games markedthe third time the national games were held in theUnited States. The first was at Iowa State Universityin Ames, Iowa, and the second was at the Universityof Nebraska in Lincoln, Nebraska. In both games,events were held at one central location; therefore, anelaborate transportation system was not required. The2014 Special Olympics USA Games were hosted byNew Jersey in June 2014. More than 4,000 athletes and

coaches from all 50 states attended, and the athletesparticipated in 16 sports over seven days. Unlikeprevious games, the 2014 USA Games were spreadacross 10 locations within a 30-mile radius in one ofthe most populous and busy areas in New Jersey. Onekey challenge was to design a convenient, simple, andreliable transportation system, which would operateon schedule for thousands of people with intellectualdisabilities under a tight budget of $600,000. The totalbudget for the entire event was $15 million.

As with any sporting mega event of this magnitude,transportation and on time attendance were among thedominant factors to the success of the games (Beis et al.2006). The 2008 Beijing Olympic Games avoided thetransportation issues by constructing new accommoda-tion and sporting facilities for most games in one centrallocation at a cost of $44 billion (Pravda.ru 2008). With ameager budget of $15 million, new construction wasimpossible for the 2014 Special Olympic USA Games,and the games needed to be spread out to existingfacilities. Thus, efficient bus routes and schedules hadto be developed to ensure timeliness and convenience,

1

Dow

nloa

ded

from

info

rms.

org

by [

108.

50.1

79.1

68]

on 1

9 A

pril

2016

, at 1

8:23

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Johnson, Zhao, and Xu: Planning and Scheduling for Special Olympics Games2 Interfaces, Articles in Advance, pp. 1–13, © 2016 INFORMS

considering 10 venues, four airports, and five specialevent locations, under a tight budget, and meeting thefollowing special requirements.

• When serving individuals with intellectual dis-abilities, consideration must be given to avoiding longwait and travel times, and minimizing the number ofbus transfers among routes; therefore, convenience hasa significant role in transportation planning.

• The transportation system (e.g., bus routes) mustbe simple, easy to understand, cost efficient, and con-venient. The system should also allow an outsideconstituent, such as a bus company, to easily executeand modify schedules, if needed.

• The system must be sufficiently reliable to handleunexpected events (e.g., random traffic patterns anddisruptions) under a reasonable budget. This one-timemega event did not allow us to make any errors (fromwhich we could learn during the event); thus, we hadto plan ahead for any unexpected issues.

Commissioned by the game organizing committee(GOC), we designed and successfully implementeda transportation system that met these requirements.The system achieved 100 percent on time performanceand 100 percent customer satisfaction, and it was onbudget despite unexpected changes in participants’travel habits; thus, it contributed to the success of the2014 USA Games. The system included both dedicatedservices for passengers whose timings and destinationswere known, and shuttle-bus services to pick up ran-dom intermittent flows of passengers. The shuttle-busservices were costly and challenging; therefore, wedeveloped an array of planning and scheduling modelsand tools ranging from demand-volume estimation andbus routing to driver scheduling, to strike a balancebetween effectiveness and simplicity for people withintellectual disabilities. These models and tools can beuseful for future Special Olympics Games and otherone-time mega-type events with dispersed locationsand under a tight budget.

Saturday Sunday Monday Tuesday Wednesday Thursday FridaySchedule 6/14/14 6/15/14 6/16/14 6/17/14 6/18/14 6/19/14 6/20/14

Games No Training Yes Yes Yes Yes Yes

Special events Welcome Opening Baseball, Olympic Baseball, Olympic Baseball, Olympic Baseball, Olympic Closingfestival ceremony Town, Cruise Town, Cruise Town, Cruise Town, Cruise ceremony

Table 1: The competition and special events were spread over seven days for this event.

Problem Description andRelationship to LiteratureThe 2014 Special Olympic USA games not only pro-vided venues for the sports and competitions, but alsoabundant special events for education and family fun.Table 1 shows a high-level view of the schedule.

Problem DescriptionThe athletes and their coaches were disseminated tomultiple locations for housing and the competitions.The College of New Jersey (TCNJ) and Rider University(RU) were the primary locations for housing (hubs).The games were hosted at TCNJ and RU, hubs 0 and 1,respectively, and eight other locations throughout NewJersey (Figure 1).

Table 2 shows the attendance for the games in eachvenue. Because of the multiple locations used forhousing, competition, and special events, the GOCneeded to provide transportation services to the athletesand coaches for the weeklong event. Transportationservices can be divided into two categories: dedicatedand shuttle bus services (Figure 2).

1. The dedicated services were provided to guaranteethe pickup and delivery of athletes and coaches toevents whose timings and destinations were known,such as opening and closing ceremonies, dinner cruises,and morning venues (every weekday from 6:30am to10:00am). These events had known headcount andstart and end times, allowing prearranged set times fortransportation to and from events. Such services arealso provided for those venues that are beyond the30-mile threshold. The objective of such services wastimeliness and reliability.

2. The shuttle services operated continuously inloops to pick up random intermittent flows, such

Dow

nloa

ded

from

info

rms.

org

by [

108.

50.1

79.1

68]

on 1

9 A

pril

2016

, at 1

8:23

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Johnson, Zhao, and Xu: Planning and Scheduling for Special Olympics GamesInterfaces, Articles in Advance, pp. 1–13, © 2016 INFORMS 3

Venue by sport

The College of NJ

Rider University

Lawrenceville School

Hun School

Princeton University

Mercer Oaks Golf

Mercer County Park

Peddie School

New BrunswickBowling Zone

10 Venues16 Sports

4,141 Athletesand Coaches

Skillman Park

Figure 1: The sports were hosted at 10 venues located within a 30-mile radius and in one of the most populatedareas in New Jersey.

as returning and sightseeing athletes and coachesfrom venues, traffic between hubs and Olympic Town(an amusement park) and Trenton baseball games,and pickup and drop-off services at airports (i.e.,

Venues Attendance

The College of New Jersey (TCNJ) 643Rider University (RU) 427Princeton University (PU) 945The Lawrenceville School (TLS) 226Mercer County Park (MCP) 952Mercer Oaks Golf (MOG) 292Skillman Park (SKM) 74Hun School (HUN) 174Peddie School (PED) 47Carolier Bowling Zone (CBZ) 361

Table 2: Each venue has a predetermined attendance based on the sportshosted.

Newark Liberty International, Philadelphia Interna-tional, Trenton-Mercer). The objective was convenience(e.g., short wait and travel time, minimum number ofbus transfers), ease of understanding, and reliability.

The planning of the dedicated services, which wasbased on the well-known taxicab problem (Giveen 1963),is relatively straightforward. The following equationshows the basic economics of dedicated services:

Number of buses =Number of riders

Bus capacity0

The most critical decision is the departure timefor the dedicated services. Because some locationswere 30 miles from the hubs and the region is highlycongested, travel could take up to 40 minutes. An addi-tional complexity was that 38 percent of competitorswere part of team competitions. To ensure adequatecapacity for this type of demand, a sufficient number

Dow

nloa

ded

from

info

rms.

org

by [

108.

50.1

79.1

68]

on 1

9 A

pril

2016

, at 1

8:23

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Johnson, Zhao, and Xu: Planning and Scheduling for Special Olympics Games4 Interfaces, Articles in Advance, pp. 1–13, © 2016 INFORMS

Dedicatedservice

Shuttleservice

Special events: Opening andclosing ceremonies,

dinner cruise

Special events: OlympicTown, Trenton baseball game

Airport: Pickup and drop offs

Venues: Sightseeing andreturning to accommodation

Venues: Morning first move-in

Figure 2: We classified the transportation services into two categories: dedicated and shuttle services. Thededicated services had known timings, destinations, and headcounts and were used for time-critical events, suchas opening and closing ceremonies and morning first move-in (i.e., morning venues). The shuttle services operatedin loops to pick up random intermittent flows of people and were used for travel to and from Olympic Town, theTrenton baseball game, airports, and sightseeing.

of buses needed to be available to transport entireteams.

The shuttle services, particularly for the returningand sightseeing flows (from 10am to 6pm each week-day), was the most complex and costly part of thegames’ transportation system. The basic economics ofa shuttle system is as follows: Given the time intervalbetween consecutive shuttles and a route (loop), thenumber of shuttles can be calculated as follows:

Number of shuttles ≥Round trip time

Time interval0

Each shuttle should have enough buses to transportevery athlete and coach who arrives during the 20-minute time interval.

Number of buses per shuttle

≥Time interval × traffic volume

Bus capacity0

Although calculating the shuttles and buses was sim-ple once the loops (bus routes) were given, determiningthe loops and the driver schedule to minimize the

passenger travel time and bus transfers, consideringthe budget constraint, was challenging. This is true forseveral reasons:

• The passenger volume was random and difficultto estimate.

• The routes had to include many locations(10 venues) and enormous combinations of loops, andscheduling the crew (the drivers) was complex.

• The system required a delicate balance among mul-tiple performance metrics (e.g., travel time, transfers,cost efficiency, and simplicity).

Relationship to LiteraturePlanning shuttle services includes traffic estimation,bus routing, and driver scheduling. It is closely relatedto the literature on public transportation planning andtransportation-services planning for special events,which includes volume estimation, route design, andvehicle and crew scheduling (Ceder 2011).

The literature on travel-volume estimation usu-ally empirically estimates origination-destination (OD)demand matrices using sophisticated statistics methods

Dow

nloa

ded

from

info

rms.

org

by [

108.

50.1

79.1

68]

on 1

9 A

pril

2016

, at 1

8:23

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Johnson, Zhao, and Xu: Planning and Scheduling for Special Olympics GamesInterfaces, Articles in Advance, pp. 1–13, © 2016 INFORMS 5

or stochastic processes based on historical traffic data(Carey et al. 1981), traffic counts and survey data (Loet al. 1999, Cascetta 1984), or population-choice modelsand behavior theory (McFadden 1974). Route designis often modeled as a variation of the vehicle routingproblem, addressing issues of travel time, waitingtime, and transfers. Sophisticated optimization mod-els and metaheuristics are developed and applied tolarge-scale public transit systems (Guihaire and Hao2008, Desaulniers and Hickman 2007). The literature onscheduling problems that address vehicles and crewsand drivers is abundant. The former assigns vehicles totrips in the transportation network (Bunte and Kliewer2009); the latter determines staff schedules to meetdemand in a cost-effective way (Ernst et al. 2004, Wrenand Rousseau 1993).

Transportation in general can be complex, and plan-ning for it is difficult (Desaulniers and Hickman 2007,Guihaire and Hao 2008); planning an effective trans-portation system for one-time mega events, such asthe Olympics, can be more challenging (Beis et al.2006). Frantzeskakis and Frantzeskakis (2006) documentthe transportation and traffic planning for the 2004Athens Olympic Games. They describe the methodsand software used to determine vehicle schedulesfor committees, sponsors, and media, based on gameschedules and shuttle buses for spectators. The keyinsights obtained (for both the Athens and BeijingGames) is that spectator movements can be served wellby a properly organized public transportation system.

Transportation services for the 2014 Special OlympicsUSA games were provided for athletes and coaches,had unique features, and required customized andnovel solutions.

• Travel volume estimation. All previous SpecialOlympics USA games were held in one central loca-tion; therefore, large-scale transportation services werenot needed. Thus, no historical data on participanttravel patterns are available and we had no lessonson transportation services for Special Olympics fromwhich we could learn. Without historical data, mostsophisticated models in the literature cannot be applied,and estimating the sightseeing volume is the biggestchallenge. Because of resource constraints, we werenot able to conduct a large-scale sampling amongparticipants; the exception was a simple survey, whichwe did among GOC members, about their likelihood

of sightseeing. Utilizing the survey and the attendanceand accommodation data, we built a simple modelto detect the travel patterns among the venues andprovided input and insights for efficient bus routingand scheduling.

• Shuttle routing. The games were characterized bya tight budget (funded by charity) and a high service-level commitment (required for people with intellectualdisabilities). The service-level metrics included bothconventional measures, such as transfers, travel, andwait times, and unconventional measures barely studiedin the literature, such as the ease of understanding(simplicity). A shuttle loop with more than two stopscould be regarded by GOC as too complex to follow.We developed a customized genetic algorithm (GA) toevaluate and optimize the shuttle loop. Inspired by theinsights obtained from our volume estimation, we cameup with a novel solution that provided direct shuttleservices among venues with the highest travel volume,considering our budget constraints. This solution metall service-level metrics.

• Bus driver scheduling. Because the system hadto serve people with intellectual disabilities, GOCmandated that the bus schedule match the erraticdemand on each route. This requirement resulted in anew hybrid shuttle-taxi model, in which we had toaccount for both bus round-trip time, as in the bus-scheduling literature, and the driver-time constraints,as in the crew- and workforce-scheduling literature.

Mathematical ModelsIn this section we focus on the shuttle services forthe returning and sightseeing flows. We first builta travel-volume model to estimate the athlete andcoach flows among the hubs and venues by day. Wethen evaluated and optimized the shuttle-bus loops todetermine the sequence of venues to visit in each loop,while keeping in mind the unique challenge (i.e., easeof understanding) that the Special Olympic Gamespresent. Finally, we designed the bus driver scheduleto determine the number of bus drivers needed pershift on each route to meet hourly demand variationover a day.

Daily Volume EstimationLittle historical data were available to allow us toresearch or to know, with confidence, the travel habits

Dow

nloa

ded

from

info

rms.

org

by [

108.

50.1

79.1

68]

on 1

9 A

pril

2016

, at 1

8:23

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Johnson, Zhao, and Xu: Planning and Scheduling for Special Olympics Games6 Interfaces, Articles in Advance, pp. 1–13, © 2016 INFORMS

of athletes and coaches; therefore, we conducted asurvey (Bixby et al. 2006) among the GOC members(e.g., chief operating officer and other members ofthe leadership team from the Special Olympics NorthAmerica office), requesting the travel habits of the ath-letes and coaches from previous games; we particularlywanted to know their habits related to traveling toeither a hub or nonhub (for sightseeing) once theirindividual competitions were completed. The survey,which took advantage of their combined experiences inorganizing international and national games, indicatedthat 25 percent of participants would sightsee aftertheir competitions. Based on the survey, we made thefollowing assumptions:

• Everyone visits at most one other venue for sight-seeing before returning to the hubs.

• The volume (of sightseeing) from one venue toanother is proportional to attendance in both venues.

• Only a fraction of participants would travel be-tween venues for sightseeing, and some participantslive at hub 0 and others live at hub 1.

Given the number of attendees at all venues andtheir travel habits (e.g., probability of sightseeing),we split the daily returning and sightseeing travelvolume among the venues and hubs into three parts:(1) sightseeing volume, (2) hub-returning volume with-out sightseeing, and (3) hub-returning volume aftersightseeing. Clearly, the volume between two venuesdepends on whether the destination venue is a hub. If itis not, then the travel volume is only the sightseeingvolume; otherwise, it is the summation of all the threeparts. We refer the reader to Appendix A for moredetails.

Through a numerical study and robust analysis(by perturbing the probability of sightseeing), weconstantly observed low volume among nonhub venuesand from hub to nonhub venues, but significantlyhigher volume among hubs and from nonhub venuesto hubs. Additional surveys indicated that athletesand coaches might prefer to return to the hubs fora change of clothes before going sightseeing, whichfurther strengthened our observations.

Bus RoutingTo design an efficient and convenient shuttle-bus systemto cover all hub and nonhub venues, we first developeda time matrix and then optimized the shuttle loops

using the volume estimation we obtained in the DailyVolume Estimation section.

We initially constructed models using Google Maps®

to collect travel times; these models served as a basisof the price analysis for bidding purposes. Six monthsprior to the event, the GOC and transportation man-agers developed an accurate representation of traveltimes by driving to and from each hub and nonhubvenue using a motor coach bus.

For bus routing we determined the number of routes(loops), the venues covered by each route, and thesequence in which they would be visited. Becauseparticipants eventually return to hubs, all routes includeat least one hub. We designed the routes to meet twoconditions:

1. Completeness: Each venue must be included in atleast one loop.

2. Nonrepetitiveness: One loop can visit a venue atmost once.

The first condition ensures all venues are connectedand the second removes redundancy in each loop. Theobjective of the shuttle-loop system is to minimize theweighted-average travel time and the number of bustransfers, considering our budget. The GOC mandatedthat the buses should be made available at each location(stop) every 20 minutes. Appendix A provides moredetails.

To build an efficient transportation network, weinvestigated multiple research streams in the vehicleand shuttle-bus routing problems. The bus transferpresented a technical challenge, because it rendersinteger programming models such as those with flowvariables (Meng and Zhou 2014) inappropriate. Tohandle this challenge and the enormous complexity ofthe problem (the possible combinations of the routesand sequences are vast), we used metaheuristics—the GA. Nia et al. (2011) and Baker and Ayechew(2003) discuss GA applications in related problems.We designed a GA (Appendix B) to determine the setof routes to cover all venues, while optimizing thevolume-weighted average travel time, the cost, and thenumber of bus transfers. Our testing showed that ourGA found a high-quality solution more quickly thanenumeration.

Using the GA and enumeration (Yih-Long andSullivan 1990) whenever computationally feasible, we

Dow

nloa

ded

from

info

rms.

org

by [

108.

50.1

79.1

68]

on 1

9 A

pril

2016

, at 1

8:23

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Johnson, Zhao, and Xu: Planning and Scheduling for Special Olympics GamesInterfaces, Articles in Advance, pp. 1–13, © 2016 INFORMS 7

evaluated many options based on metrics of cost effi-ciency, convenience, and simplicity. The followingoptions are representative.

1. Direct pairs: This simple option connects each pairof venues by a direct loop. This could be an expensiveoption, because the volume is disaggregated to eachpair; however, it provides the maximum conveniencewith the shortest travel time and no bus transfers.

2. One loop for all venues: This simple option con-nects all venues by a single route. It aggregates thevolumes on all pairs of venues to save money andrequires no bus transfers. However, it may requirelengthy travel times, because riders may need to makemany stops before reaching their destination.

3. Two loops with seven venues: All loops includethe two hubs and five other venues. This option ismore complex, because the participants must memorizewhich loop serves which seven venues. The travel timeand cost are between those in option 1 and option 2,and this option may require one bus transfer.

4. Three loops with seven venues: Each loop servesthe two hubs and five other venues. It may decreasethe travel time in comparison to option 3; however, itis harder to follow and may increase the number ofbus transfers.

5. Two hubs and one nonhub: A loop covers the twohubs and each nonhub venue. It is simple to followand ideal for the returning flows from nonhub venuesto hubs; however, a sightseer at a nonhub venue mayneed to transfer to another bus at least once.

6. Direct hub and nonhub: A loop directly connectseach hub and each nonhub venue. It is more expensivethan option 5, but it results in shorter travel times andis easy to follow.

Table 3 shows the results from these options, includ-ing required budget and average travel times.

After we showed the results to the leadership team,GOC picked option 6 (direct hub and nonhub routes)based on its convenience, shorter travel times, andsimplicity of comprehension for individuals with intel-lectual disabilities. The relatively small volume ofsightseeing flow and the survey referencing athletesand coaches returning to hubs before going sightseeingeased the bus-transfer concern. Option 6 also met thebudget requirement.

Average travelOption Budget ($) time (minutes)

1. Direct pairs 418,545 162. One loop for all venues 100,800 683. Two loops with seven venues 144,000 324. Three loops with seven venues 136,800 245. Two hubs and one nonhub 142,335 256. Direct hub and nonhub 153,945 23

Table 3: Our genetic algorithm, which we labeled SO2014, computed therequired budget and average travel time for each option and found thatoption 6 (direct hub and nonhub) provided the shortest average travel timewithout exceeding the budget.

Bus Driver SchedulingTo schedule bus drivers, we first disaggregated thedaily volume into hourly demand, which we definedas the number of people needing transport per unitof time (hour). To this end, we had to consider thegame schedule and lunch schedule; lunch is providedat each venue. We expected that participants of a gamewould stay until the game was over and then return tothe hubs before sightseeing. In addition, if the gamecompleted before lunch, people would leave after lunch.Finally, people might go sightseeing at other venues,and the timing would occur equally likely from thetime they return to the hub until 5pm. The estimatedhourly volume typically peaked right after lunch anddiminished toward the end of the day (6pm).

Given the bus routes and the hourly volume esti-mates by day, we needed to determine the bus driverhourly schedule and the number of drivers needed pershift (Yoshitomi 2002). Each bus would depart fromhub 0 or 1 to a nonhub venue and return to the samehub. Each bus would only serve one venue, and thenumber of buses used in each would vary over time.

We broke down hourly volume on each route into 15-minute time buckets (the actual time interval for mostloops) and developed a mathematical programmingmodel to determine which driver would take whichroute during which time bucket to meet changingdemands over a day at the lowest bus driver cost. Inaddition to the demand constraints, we had to ensurethat a bus driver works at least a specified numberof consecutive hours (a typical time constraint forcrew- scheduling problems), and that the bus driveris not available to serve a venue again until after theround-trip (a typical constraint for bus-scheduling

Dow

nloa

ded

from

info

rms.

org

by [

108.

50.1

79.1

68]

on 1

9 A

pril

2016

, at 1

8:23

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Johnson, Zhao, and Xu: Planning and Scheduling for Special Olympics Games8 Interfaces, Articles in Advance, pp. 1–13, © 2016 INFORMS

Phase Time frame Structure GOC feedback

1 Two years prior to games: Gameschedule 60% complete

Shuttle-bus routes Lack of travel habits, complex routes,not intuitive

2 Six months prior to games: Gameschedule 95% complete

Includes travel habits to fine-tune volumeestimation model, simpler bus routes (e.g.,Options 5 and 6)

Option 6 selected based onconvenience and ease ofunderstanding

3 One month prior to games: Gameschedule complete

Breaks down daily volume into hourly volume,and schedules bus drivers on each route

Adopted, enhanced byrisk-management plans

Table 4: We implemented a three-phase approach, interacted repeatedly with the games organizing committeemembers to obtain the latest information and their feedback, and addressed their requests with new solutions.

problems). Appendix B provides the detailed model for-mulation. We implemented the model using the Pythonprogramming language and the Gurobi optimizationprogram.

ImplementationThe implementation process was challenging for severalreasons: (1) much of the information that was criticalwas available to us gradually (Varelas et al. 2013);(2) many modifications were necessary throughout theplanning phase (Bixby et al. 2006), such as fluctuatedgames budgets, game schedules, and venue adjust-ments; and (3) we had to prepare for numerous layersof risks and contingencies.

Implementation PhasesWe used an iterative implementation process and athree-phase approach (Table 4). In phase 1, approxi-mately two years prior to the games, we estimatedtraffic volumes based on attendance at game venuesand created shuttle-bus routes based on daily traffic vol-ume, thus laying the foundation for our bid. In phase 2,about six months prior to the games, we conducted asurvey and included the athlete travel-habit matrix,which significantly improved our volume estimation.We also studied simpler route options—options 5 and6 in the Bus Routing section. Phase 3 incorporated theGOC leadership team’s selection of the simple directhub to nonhub routes (option 6) and scheduled thebus drivers to meet hourly demand.

Bus CompaniesTransport for the games was delivered through apartnership between the GOC and the Academy andFirst Student bus companies. Table 5 shows the vehicle

Type Capacity Cost/Hour ($) Total available

Motor coach 50 90 300Low step 50 75 10School bus 45 54 100

Table 5: Two New Jersey-based bus companies provided an estimate on thenumber of buses available, the capacity, and the cost per hour for each typeof bus.

types and capacities. The motor coach bus was usedfor competition, airport services, and special events,and the low-step transit bus was used for the shuttleservices from the TCNJ and Rider venues. The schoolbus was used for the evening showcase sports andspecial events. By contract, the minimum time requiredto engage a bus was four hours for all services.

Once the competition and special event scheduleswere solidified and provided to the state delegations,Academy, and First Student, the GOC conducted a table-top exercise to determine the number of buses needed.This proved to be very fruitful because it enabled us toidentify potential problem areas and assign the numberof drivers needed throughout each day.

Both transportation companies used the plan toschedule routes and the number of buses needed totransport the athletes and coaches to their respectivecompetition venues and return them to their originatinghubs. This information was critical for determiningthe number of buses they would need to add to theirNew Jersey fleets, because they provide services forthe entire northeast region.

Risk-Management and Contingency PlanOne challenge in hosting a mega event is that thoseresponsible must do everything correctly the firsttime, because such an event provides no opportunity

Dow

nloa

ded

from

info

rms.

org

by [

108.

50.1

79.1

68]

on 1

9 A

pril

2016

, at 1

8:23

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Johnson, Zhao, and Xu: Planning and Scheduling for Special Olympics GamesInterfaces, Articles in Advance, pp. 1–13, © 2016 INFORMS 9

for learning (based on mistakes) and improvementduring the event. Consequently, we had to identifyall potential risks and contingencies and develop arobust transportation system. Therefore, prior to thegames, we put together a list of common risks for atransportation network.

1. Traffic: The venues were scattered throughoutcentral New Jersey, an area that is busy with its typicaldaily traffic. Coupled with the influx of new observers,we expected a driving-time increase of 20–50 percentduring morning rush hours relative to non-rush hours.

2. Volume: The passenger volumes for all trans-portation services were best estimates based on gameschedules and our understanding of travel habits.Overestimating can result in waste; however, underesti-mating can result in long wait times and poor customerservice.

3. Technology: Various communication devices, suchas radios, global positioning systems, cellular phones,and handheld tablets, which we would use in ourtransportation system, can malfunction or break down.

4. Weather: This event would be held during thesummer; therefore, we would have to consider the prob-ability of excessive heat and thunderstorms, possiblyresulting in delays.

5. Driver error: The drivers are professional driversfrom well-established bus companies; however, becauseof human error, some drivers might not arrive at thedesignated locations on time.

6. Route closures: Scheduled outages or unplannedaccidents are always a possibility. Alternate routeswould need to be identified and scheduled at theplanning stage.

7. Mechanical: Vehicle performance, even withproper maintenance, can be affected by unplannedmechanical problems.

In response, we developed the following risk-management strategies. First, we reserved a smallbudget to enhance the reliability of the transportationsystem by pooling a fleet of extra buses at a few strategiclocations to provide emergency services; this provedto be valuable during the games, because our hourlytraffic prediction did not precisely match the peakdemand. Second, the GOC developed a transportationrisk-management plan with the assistance of the buscompany operations manager. To ensure communica-tion was cohesive throughout the multiple locations,

a main operations center (MOC) was established atone of the campuses. The MOC, which representedseveral departments—including games transportation,the New Jersey Department of Transportation (NJDOT),law enforcement, and public safety—operated on a24-hour schedule. To address transportation issues, theMOC contained multiple display screens connected toreal-time global positioning system-monitoring andthe NJDOT regional In Transit Visibility (ITV) sys-tems. The benefit of using real-time and ITV monitoringwas the ability it provided to track every bus in thenetwork and the option to use alternate routes if needed.

Results and AchievementsResultsSpecial Olympics 2014 was the first USA game heldat dispersed locations in a populous area such thatit required large-scale transportation services. Ourmodels and algorithms, together with an iterativeplanning and implementation process, resulted in atransportation system that was timely, convenient, costefficient, easy to understand, and reliable. First, itprovided a convenient 20-minute interval betweenconsecutive buses and an average travel time of about23 minutes over all pairs of locations. Second, thesystem provided easy-to-understand, one-stop serviceswithout the transfers; at nonhub venues, participantsneeded only to follow the color of a sign on the bus(i.e., red for RU and blue for TCNJ). Finally, it wascost efficient and reliable. It met the budget constraintsand had a sizable surplus of $45,000 left over, thusallowing us to plan for contingencies.

The transportation system we developed may beapplied to other one-time mega events held in sparselocations, such as Olympic Games, mega exhibitions,and conferences.

AchievementsDespite unexpected travel-habit changes because ofNew Jersey’s many attractions, the transportation sys-tem implemented for the 2014 Special Olympic USAGames was a great success. The system (1) was onbudget; (2) achieved 100 percent on-time performanceat competition, special events, and airports; (3) main-tained an average of 20-minute intervals to all venues,as planned; and (4) achieved 100 percent customer

Dow

nloa

ded

from

info

rms.

org

by [

108.

50.1

79.1

68]

on 1

9 A

pril

2016

, at 1

8:23

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Johnson, Zhao, and Xu: Planning and Scheduling for Special Olympics Games10 Interfaces, Articles in Advance, pp. 1–13, © 2016 INFORMS

satisfaction, based on random interviews of 20 athletesand coaches and observations by several hundredoperation managers at bus stops.

As one coach we interviewed said: “Couldn’t haveasked for a better bus transportation system. Buseswere on time and didn’t have to wait.”

ImplicationsThe study provides an approach that allows an eventhost to do more with less for mega events such asOlympics. The norm of the current practice is heavilygeared toward the construction of large-scale multi-purpose facilities (e.g., 2008 Beijing Olympic Games)to avoid a nightmarish transportation problem. Theconstruction of such facilities, however, often results inlarge capital expenditures upfront and underutilizationafter the event. To avoid this problem, one can utilizeexisting facilities that may, however, be sparsely located.The success of the transportation system developed forthe 2014 Special Olympics USA Games demonstratesthe transportation feasibility for holding such megaevents at dispersed locations in a highly populous andcongested area under a tight budget. This is especiallyrelevant because the transportation services were pro-vided to individuals with intellectual disabilities whorequired high service levels.

ConclusionsIn this paper we discuss our development of a trans-portation system for Special Olympic Games, whichstrikes a balance between simplicity and effectiveness.Shuttle-bus service was the most costly and challengingcomponent. To address it, we developed models andalgorithms for (1) volume estimation for sightseeingdemand, with minimum available data; (2) bus routingthat is convenient, cost effective, and easy to under-stand; and (3) bus driver schedules that incorporate keyfeatures of both bus- and crew-scheduling problems.

Our experience with this project provided many valu-able lessons. First, the key challenge to make a real-lifeimpact was not only the model and algorithm develop-ment, but also the implementation, specifically, strikinga balance between effectiveness and simplicity. Second,one-time mega events allow no errors during the event;thus, one must plan ahead for any unexpected issues.For example, our hourly volume estimation missed thepeak demand because of unexpected changes in the

participants’ travel habits. Fortunately, we reserved abuffer of buses, which could be quickly put into service.Third, necessary information was not always availableat the start of the project, and an iterative process offrequent interactions with customers was necessary.Finally, the game and special event schedule heavilyaffected the efficiency of the transportation system.Thus, one of our planned future studies on such megaevents is to jointly optimize the event schedule andtransportation systems.

AcknowledgmentsThe second author was supported in part by the NationalScience Foundation [Grants 0747779 and 1537591]. The secondauthor was also supported in part by National NaturalScience Foundation of China [No. 71531003]. This work wasdone under the auspices of the Supply Chain AnalyticsLaboratory (SCAL) at Rutgers University.

Appendix A: Weighted-Average Travel Time

Travel TimeGiven the travel time between venue i and j , tij , the weighted-average traveling time is

∑

4m1n54Vmn ×∑

4i1 j5 tijymnij 5

∑

4m1n5 Vmn

1

where 4m1n5 refers to a pair of origins and destinations, ymnij

indicates whether 4i1 j5 is on the shortest path from m to n(depending on the routes of the shuttle-bus system), and∑

4i1 j5 tijymnij is the shortest time to travel from venue m to n.

If a bus switch occurs on the shortest path, a time penalty isincurred to represent inconvenience.

Appendix B: Transportation Models

Volume Estimation ModelsAssumptions

• Everyone visits at most one other venue for sightseeingbefore returning to the hubs.

• Between venues i and j , the volume (of sightseeing)from i to j is proportional to attendance in i and attendancein j .

• The fraction of participants who would travel betweenvenues for sightseeing is �= 25%. The fraction of participantsliving at hub 0 (1) is �0 = 50% (�1 = 50%), respectively.

Formulation2 Parameter and indices:

Ai Attendance at each venue, i = 0111 0 0 0 19.Sij Sightseeing volume from i to j (Sij5.Hij Hub-returning volume without sightseeing from i to j

(where j is a hub).Rij Hub-returning volume to j (a hub) after sightseeing at i.

Dow

nloa

ded

from

info

rms.

org

by [

108.

50.1

79.1

68]

on 1

9 A

pril

2016

, at 1

8:23

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Johnson, Zhao, and Xu: Planning and Scheduling for Special Olympics GamesInterfaces, Articles in Advance, pp. 1–13, © 2016 INFORMS 11

Given origin venue i, we discuss two scenarios about thedestination venue j :

• Case 1: Venue j is not a hub; sightseeing volume occursonly from i to j where Sij = �Ai × 4Aj/

∑

k 6=iAk5 Thus, thetravel volume between i and j is Vij = Sij .

• Case 2: Venue j is a hub; sightseeing volume occurs fromi to j , Sij ; hub-returning volume from i without sightseeing,Hij = �j 41 −�5Ai; and hub-returning volume after sightseeingat venue i, Rij = �j

∑

k Ski. Thus, the travel volume between iand j is Vij = Sij +Hij +Rij .

Bus Routing: Genetic Algorithm (GA)The algorithm, SO2014 GA, includes five iterations: initiation,evaluation, condition check, crossover, and mutation (Fig-ure B.1). We defined the sequence in five steps with the fifthstep returning to Step 2. The CONDITION defined in theprogram was the number of iterations. Once the numberof iterations was reached, the CONDITION was met. Incomputation, we assumed that each stop takes a maximum offive minutes and each switch between loops cost 20 minutes.We obtained the hourly volume by evenly spreading out thedaily volume to the periods available.

Our numerical study showed that enumeration was notcomputationally feasible for two or more loops. Too manycombinations were generated, and the computation timesneeded to arrive at an optimal solution proved to be unreal-istic and would not be transferable to future events. For such

Step 1

Step 3

Step 4

Step 5

Step 2

Conditionmet?

Initialization

Randomly generates Nsolutions to form firstpopulation

Calculate traveling timebetween venues, ifswitching bus occurs,append time penalty

Find the shortest pathbetween each pair of venuesTake average

Introduces random changes

Reintroduces geneticdiversity back into population

Fitness assignment

Proportional-Probabilities(Roulette)

Two parents combined toform offspring

Evalution

Yes

End Crossover

Mutation

Figure B.1: This flowchart of our SO2014 genetic algorithm shows its five iterative steps.

options, we used SO2014 GA. We carried out all numericaltests on an Intel Core-i7 CPU, 3.5 GHz (8 core) desktopworkstation with 16 MB of memory. The algorithms werecoded in Python and run using the Gurobi optimizationprogram.

Bus Driver Scheduling ModelsModel Formulations2 Model Development for Phases 2 and 3

This model considers time windows for buses. We expectthat a venue will have bus arrivals every 15 minutes. From10:00am to 6:00pm, we divide the time frame into 32 sections:10:00–10:15am, 10:15–10:30am, 0 0 0 , 5:45–6:00pm. For example,suppose the time horizon begins at 10:00am (so that the firstbus will be able to drive 30 minutes to arrive at the farthestvenue). We convert 10:00 to A0 = 60; 10:15 to A1 = 75, and soon. In general, Ak = 60 + 15 · k, k = 0111 0 0 0 132.

We hope a bus would serve only one venue during a dayto reduce the risk. By this assumption, the problem can bedecomposed into a set of subproblems, with each subproblemconsidering the bus-scheduling problem for one venue only.

Parameters:

C: Capacity of a bus.Vk: Volume at cycle k.Ak: Arrival time of cycle k.T : One-way travel time from hub to venue.

Cycle duration: 15 minutes

Dow

nloa

ded

from

info

rms.

org

by [

108.

50.1

79.1

68]

on 1

9 A

pril

2016

, at 1

8:23

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Johnson, Zhao, and Xu: Planning and Scheduling for Special Olympics Games12 Interfaces, Articles in Advance, pp. 1–13, © 2016 INFORMS

Variables:

xbk: Binary, when =1, bus b will satisfy the demand of thekth cycle.

yb : Binary, when =1, bus b is used.ub : Integer, the last cycle to satisfy.lb : Integer, the first cycle to satisfy.hb : Integer, reservation length for bus b (hours).

Objective Function:min

∑

b

hb (B1)

Constraints:The volume at each cycle must be satisfied (capacity con-straint):

C∑

b

xbh ≥ Vk for all k (B2)

1. If bus b visits venue at Ak (cycle k), the time frameshould satisfy (nonoverlapping task constraint):

xbk4Ak − 2T 5+M · 41 − xbk5≥ xbiAi for all i ≤ k− 1 (B3)

2. Relationship between xbk and yb (big-M for bus):∑

k

xbk ≤M · yb for all b (B4)

3. A bus, if reserved, must be reserved for at least fourhours (minimum-time requirement)

ub− lb +M41 − yb5+ 2T ≥ 4 · 60 for all b (B5)

4. Relationship among ub , lb , and hb (starting and endingtime constraints):

ub and lb (B6)

lb ≥ xbkAk for all k (B7)

lb ≤ xbkAk + 41 − yb5M + 41 − xbk5M for all k (B8)

ub≥ lb (B9)

ub− lb + 2T ≤ 60 ·hb

+M41 − yb50 (B10)

ReferencesBaker BM, Ayechew MA (2003) A genetic algorithm for the vehicle

routing problem. Comput. Oper. Res. 30(5):787–800.Beis DA, Loucopoulos P, Pyrgiotis Y, Zografos KG (2006) PLATO

helps Athens win gold: Olympic Games knowledge modelingfor organizational change and resource management. Interfaces36(1):26–42.

Bixby A, Downs B, Self M (2006) A scheduling and capable-to-promise application for Swift & Company. Interfaces 36(1):69–86.

Bunte S, Kliewer N (2009) An overview on vehicle scheduling models.Public Transport 1(4):299–317.

Carey M, Hendrickson C, Siddharthan K (1981) A method for directestimation of origin/destination trip matrices. Transportation Sci.15(1):32–49.

Cascetta E (1984) Estimation of trip matrices from traffic counts andsurvey data: A generalized least squares estimator. TransportationRes. Part B: Methodological 188(4–5):289–299.

Ceder A (2011) Public-transport vehicle scheduling with multi vehicletype. Transportation Res. Part C: Emerging Tech. 19(3):485–497.

Desaulniers G, Hickman M (2007) Public transit. Barnhart C,Laporte G, eds. Handbooks in Operations Research and ManagementScience (Elsevier, London), 69–127.

Ernst AT, Jiang H, Krishnamoorthy M, Sier D (2004) Staff schedulingand rostering: A review of applications, methods and models.Eur. J. Oper. Res. 153(1):3–27.

Frantzeskakis J, Frantzeskakis M (2006) Athens 2004 Olympic Games:Transportation planning simulation and traffic management.Inst. Transportation Engrg. J. 76(10):26–32.

Giveen SM (1963) A taxicab problem with time-dependent arrivalrates. SIAM J. Comput. 5(2):119–127.

Guihaire V, Hao J-K (2008) Transit network design and schedul-ing: A global review. Transportation Res. Part A: Policy Practice42(10):1251–1273.

Lo HP, Zhang N, Lam WHK (1999) Decomposition algorithm forstatistical estimation of OD matrix with random link choiceproportions from traffic counts. Transportation Res. Part B: Method-ological 33(5):369–385.

McFadden D (1974) Measurement of urban travel demand. J. PublicEconom. 3(4):303–328.

Meng L, Zhou X (2014) Simultaneous train rerouting and reschedulingon an N-track network: A model reformulation with network-based cumulative flow variables. Transportation Res. Part B:Methodological 67(C):208–234.

Nia SG, Sharif B, Habibzadeh N, Rezvani M (2011) Using the geneticalgorithm for optimization of the integrated urban transportationsystems. J. Tech. Management Growing Economies 2(1):103–113.

Pravda.Ru (2008) Beijing Olympics to cost China 44 billion dollars.Accessed June 10, 2015, http://english.pravda.ru/sports/games/06-08-2008/106003-beijing_olympics-0/.

Special Olympics, Inc. (2015) Accessed June 10, 2015, http://www.specialolympics.org/mission.aspx.

Special Olympics: Our Athletes (2014) Accessed April 9, 2014,http://www.specialolympics.org/Common/Frequently-Asked-Questions.aspx.

Varelas T, Archontaki S, Dimotikalis J, Turan O, Lazakis I, Varelas O(2013) Optimizing ship routing to maximize fleet revenue atDanaos. Interfaces 43(1):37–47.

Wren A, Rousseau J-M (1993) Bus driver scheduling—An overview.Daduna JR, Branco I, Paixão JMP, eds. Computer-Aided TransitScheduling (Springer-Verlag, Berlin), 173–187.

Yih-Long C, Sullivan RS (1990) Schedule generation in a dynamic jobshop. Internat. J. Production Res. 28(1):65–74.

Yoshitomi Y (2002) A genetic algorithm approach to solving stochasticjob-shop scheduling problems. Internat. Trans. Oper. Res. 9(4):479–495.

Verification LetterLillian Narvaez, Chief Operating Officer, 2014 USA Games,64 Walt Whitman Way, Hamilton, New Jersey 08690, writes:

“In June 2014, the Special Olympics USA Games washosted by the state of New Jersey. 3,300 athletes with intel-lectual disabilities and 1,000 coaches competed in 16 sportsacross 10 locations and over 70,000 spectators were in atten-dance. As with any sporting mega-event of this magnitude,transportation was among the top dominant factors in realiz-ing success. Efficient bus routes and schedules were needed to

Dow

nloa

ded

from

info

rms.

org

by [

108.

50.1

79.1

68]

on 1

9 A

pril

2016

, at 1

8:23

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Johnson, Zhao, and Xu: Planning and Scheduling for Special Olympics GamesInterfaces, Articles in Advance, pp. 1–13, © 2016 INFORMS 13

ensure timeliness and convenience between 10 venues, threeairports, and five special event locations while remainingunder the transportation budget of $600 K.

“We solved this transportation problem using a three-phaseapproach. Phase 1 optimizes the number of shuttle-loops andbuses required to efficiently transport athletes and coaches tocompetition venues using the enumeration method. We thendeveloped a time matrix and volume estimation models toensure the transportation network was equal to or underthe budget. During this phase, we also designed a geneticalgorithm enabling us to find the sub-optimal solution fasterthan by enumeration alone. Phase 2 sees the integration of theathletes proposed travel habits and a more focused volumeestimation model detailed in hourly variations instead of adaily volume total. Finally, Phase 3 solves the shuttle-busproblem by a more direct approach. Due to the constrainedcompetition and special event schedules, we needed a directroute to and from each hub and non-hub venue to ensure alltimelines were met.

“With the experience of the Academy Bus Company opera-tion managers, we evaluated this three-phase methodologyby applying to a real-world mega-event (the 2014 SpecialOlympics USA Games) resulting in a huge success. Eventhough this event did not result in substantial transportationsavings, we were however, able to streamline the processensuring athletes with intellectual disabilities and their coacheswere able to make every competition with zero delays.”

Andrew Johnson is a professor in the Department ofMarketing and Business Information Systems at RowanUniversity. He holds a PhD in supply chain management

and marketing sciences from Rutgers University. His generalinterest lies in supply chain management and logistical sys-tems with special focus on transportation planning, inventorycontrol, military applications, and the integration of supplychain management with project management. He teachessupply chain logistics management, business logistics, globalsupply chain management, principles in transportation, andresearch methods in marketing for undergraduate and MBAprograms at Rowan University. He is a 20-year veteran of theUnited States Air Force, retiring in 2007.

Yao Zhao is a professor in the Department of Supply ChainManagement at Rutgers University. He holds a PhD degree inindustrial engineering and management sciences from North-western University. His research interests lie in supply chainmanagement, new product development, and healthcare.He has worked on applications in aerospace, transportation,energy, healthcare, and pharmaceutical industries. He haspublished widely in leading operations research and opera-tions management journals and serves as an associate editorof Operations Research. He was the recipient of honorablemention in the M&SOM student paper competition in 2001,the National Science Foundation Career Award on Manufac-turing Enterprise Systems in 2008, and the Dean’s Award forMeritorious Research in 2011. He teaches core operations andsupply chain management courses for undergraduate, MBA,Executive MBA, and PhD programs at Rutgers University.He is a case writer and has extensive consulting experiencewith companies and governmental agencies.

Xin Xu is a senior financial consultant at Ernst & YoungLLP. He holds a PhD in supply chain management andmarketing sciences from Rutgers University.

Dow

nloa

ded

from

info

rms.

org

by [

108.

50.1

79.1

68]

on 1

9 A

pril

2016

, at 1

8:23

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.