A Survey-Based Mobilty Model of People forSimulation of Urban Mesh Networks

Jonghyun Kim Stephan Bohacekkim@eecis.udel.edu bohacek@udel.edu

Abstract– In this paper a mobility model of people inurban areas for mobile wireless network simulation is pre-sented. A 3-layer hierarchical approach is taken wherethe highest layer is an activity model that determines thehigh level activity that the node is performing (e.g., work-ing). The second level is a task model that models thespecific task within an activity (e.g., meeting with threepeople). And the third level is an agent model that deter-mines how the person moves from one location to another.These three models are based on a number of surveys anddata sources. The activity model is based on a recent USDepartment of Labor Bureau of Labor Statistics time usestudy. Such time use studies gather detailed informationabout how the interviewees spent their time. The taskmodel mostly focuses on mobility of office workers andis based on the current findings from research on meet-ings analysis. The agent model is based on the work fromurban planning that has collected extensive knowledge ofpedestrian flow. The models presented are implementedin a mobility simulator that is integrated with a wirelesspropagation model (described elsewhere).

I. Introduction

There has been growing interest in urban mesh net-works. Recently, the city of Philadelphia has entered thefinal planning stages for the deployment of a mesh net-work that intends to provide coverage to the entire 135sq. mi. city with 4000 fixed base stations [1]. Severalother cities are considering similar deployments [2].While these deployments are proceeding, there remains

a large number of issues relating to the performance ofthese networks. However, since these networks have yetto be constructed, in order to study them and validatefindings related to them, one must utilize simulation. Thesimulation of large scale urban is larger unexplored andquite different from the simulation techniques currentlyused for mobile wireless networks. For example, today,most researchers focus on small-scale mobile ad hoc net-works. The premise behind much of this work is thatsuch networks will be utilized by the military. However,urban networks such as the one proposed by Philadelphiaare different from military ad hoc networks and the net-works currently simulated in several ways. Specifically,in civilian urban networks are fixed base stations, wire-less relays, handheld terminal carried by civilians, andterminals in vehicles. Furthermore, the number of civil-ians hoped to use such networks far exceeds the numbersof nodes considered in today’s simulation of military adhoc networks. For example, in today’s literature, it is notuncommon to see simulation that consist of less than 100nodes over a 1km2 region, whereas during the lunch hour,a 1km2 region in midtown Manhattan has about 10000people outdoors [3]. Furthermore, today’s simulations of

military ad hoc networks utilize mobility models that at-tempt to capture the wide range of mobility that suchnetworks might experience. Such models do not makesense to simulate urban mesh networks where mobility ismore civil.The principle aspects of urban civilian mobility is the

pedestrians move along sidewalks and inside buildings,while vehicles move along roads. Furthermore, pedestri-ans and vehicles obey various rules. For example, vehiclesand pedestrians, more or less, obey traffic signals. Also,in order for a faster node to pass a slower node, it mustgo around the slower node. If there is no room to pass,then the faster node must decrease its speed to that ofthe slower node and follow until there is room to pass.Such rules have an impact on the distribution of nodes.For example, the passing rules result in cluster of nodesfollowing a slower node of group of nodes. Indeed, withurban planning, it is known that both vehicles and pedes-trians arrive in bursts [3].This paper presents a detailed mobility model for ur-

ban pedestrians during a workday. This model is basedon three mature research areas, urban planning [3], [4],meeting analysis [5], and time use [6]. The resultingmodel is a three layer hierarchical model. The highestlayer is the activity model that determines the high-leveltypes of activities. The activity model determines thetime when people start and end various activities. Thedata used to generate this model is from the recent USBureau of Labor Statistics (BLS) use of time study [7].Such time use studies have been actively performed forthe past forty years [6]. The 2003 US BLS study marksthe beginning of a yearly study of time use that is basedon ten years of planning within the BLS. The 2003 studyincludes interviews with roughly 20,000 people. Of those,around 5000 resided in metropolitan areas and were usedin the activity model presented here. Furthermore, theBLS determined weightings to account for over samplingof some types of people (e.g., unemployed people tend tobe at home at the time of the interview call and tend tobe over sampled). Hence, the significance of the studyexceeds the 20,000 that were actually interviewed.The second layer of the pedestrian mobility model is

the task model. Within an activity, there may be a largenumber of tasks. For example, our model focuses on of-fice workers where there are two types of tasks, workingat their desk and meeting with other workers. The signif-icance of these worker tasks is that they model the mobil-ity of nodes within buildings as well as the clustering ofnodes within buildings. The basis of this part of the mo-bility model is several seminal studies of worker meetingsperformed within the management research community

MeshNets 200510 July 2005Budapest, Hungary

93

[5].The third layer of the mobility model is the agent

model. Such agent models are investigated within thearchitecture community and define how the node navi-gates to its desire destination. We based our model onurban planning research, especially the seminal work ofPushkarev and Zupan [3] that includes findings of theirown extensive studies as well as results from several otherpedestrian mobility studies. The pedestrian chapter ofthe US Highway Capacity manual [4] is also based onthis work.The models presented here are part of a simulator de-

veloped by the authors and others. Along with mobility,the simulator includes wireless transmission propagationmodeling along with tools for designing urban maps in-cluding sidewalks, roads, buildings, types of buildings,etc. It is crucial that a simulator of urban mobility isintegrated with propagation models. For example, themobility model may specify that a node move from in-doors to outdoors. However, there are dramatic differencebetween propagation indoors and propagation outdoors.Merely modeling the mobility of the node without a rea-sonable propagation model will likely distort the conclu-sions gained from the simulation.

II. Related Work

The fact that mobility plays an important role inMANET simulation has demonstrated in [8]. The specificimportance of realistic simulation was been demonstratedin [9] and [10], These paper also demonstrated the impor-tance of integrated mobility and propagation simulation.In this paper, we only describe our mobility model withpropagation left for a companion paper.A review of mobility models of MANETs can be found

in [11]. However, MANET research has been dominatedby the random way-point mobility model. While such amodel may provide valuable insights, it is far from real-istic and will likely not give reasonable performance es-timates of urban mesh networks. In [11] and [8] mobil-ity models are present where the nodes are restricted tomove along road of hypothetical grid-like cities. In [12]the graph is not grid-like but based on a Voronoi graph.In [13], the a random graph is used. In these cases, themobility is essentially random way-point, but restrictedto a graph.Recently there has been interest in developing more de-

tailed models along the line discussed here. For example,[14] and [15] discusses a empirical model based on obser-vations of pedestrian on a university campus. While [16]describes a realistic model of MANET mobility duringdisasters.One of the most detailed mobility models is GEMM

[17]. GEMM is an agent-based model where severalfactors impact the mobility of the node. For example,GEMM includes attraction points as well as habits toinfluence the mobility. A noted drawback of this workis that realistic values of the model parameters are notknown.Urban planning, sociology, and architecture have been

activity developing mobility models for at least fifty years

Stores/gymrestaurant

Offices Homes(apartments)

Figure 1. Locations in different types of buildings. Locations aremaked with a circle while arcs are indicated by thin lines. thethick lines denote walls. The store/gym/resturant structure issuch that each third oif the layout can be any of the options.The appartment buidling shown has four apartments each withfive rooms. The size of rooms is approximately constant. Whilelarger buildings can a accomidate more rooms.

[18]. As a result, there are far too many approaches andpapers to mention here. However, for pedestrian mobility,it appears that the models compose of and activity-basedmodel and an agent-based model are the state-of-the-art[19], [20]. This paper utilizes this layered approach tomobility modeling. Furthermore, the model presentedhere is based on data collected through rigorous surveys.

III. Building and Location Types

The mobility model defines the location of nodes. tothis end, the urban region under consideration is mod-eled as a graph; all locations are modeled as vertices andpathways between the locations are arcs. Nodes can onlymove along the arcs. Locations include offices, hallways,sidewalks, crosswalks, rooms within a home, seating lo-cations in restaurants, locations within a store, and loca-tions within an gym. As shown in Figure 1, buildings aremodeled simply. Our simulation tools require manuallygenerate buildings. The interior layout is automaticallygenerated.

IV. Activity Model

This part of the mobility model is based on the USBureau of Labor Statistics 2003 time use study [7]. Thisstudy identify a large number of activities. We focus onthose activities that indicate location and group locationtogether that are done in the same location (e.g., homeactivities). While the BLS study also collected coarselocation information. Both activity and location infor-mation were used to determine the location used in themodeling effort. We focus on eight types of activities:working, eating not at work, shopping, at home, receiv-ing professional service, exercise, relaxing, and droppingoff someone. Note since we are focus on location and mo-bility, eating at work is counted as work. Eating includeseating at a restaurant and buying food some where. Shop-ping includes all types of shopping except buying food.Receiving professional service ranges from things such asgetting medical attention to receiving to household main-

94

tenance that is not performed at home1 .During the simulation initialization, each node is given

an office and home. It is assumed that work is done withinbuilding where the nodes office is (work done at home isincluded into the at home activity). Eating is done ata restaurant (eating at home is included into at homeactivity). Shopping is done at one of many stores. re-ceiving professional service is done at an office that is notthe nodes office (unless the service is provided while atwork or at home in which case the activity is included aswork or and at home activity). Our current model doesnot specify a location for relaxing and dropping some oneoff. Dropping someone off includes meeting children atschool and taking them home and taking an parent tothe doctor. For the purpose of mobility modeling, wemodel such activities as a trip home followed by a trip toan office location. The node remains at the office loca-tion until the drop off activity if complete. The relaxingactivity is modeled as going to an office location (muchlike receiving professional service).This model effort focuses on the work day which consist

of being at home, going to work, working, and perhapstaking a break and returning to work and then leavingwork and returning home. The model neglects activitiesbefore and after work. Future work will include the restof the day.For each person, the following steps are taken to deter-

mine the activities that they perform.

1. Select a home and office.2. The arrival time at work is determined.3. The duration at work is determined.4. Determine if a break from work is taken. (The next 5steps assume a break is taken.)5. The break start time is determined.6. The number of activities performed during a break isdetermined7. Which activities are performed during the break is de-termined.8. The duration of each activity is determined.9. The arrival time back at work is determined and it isdetermined whether about break is taken. If so, steps 5-9are repeated.

Selection of home and office An office for each nodeis selected at random. Once an office is selected, a homeis selected that is nearby the office. Specifically, a home isselected so that the distance from the home to the officematches the distribution shown in 7. This distribution isbased on walking distances observed by Pushkarev andZupan. The model also allows for nodes to not walk towork, but to arrive via the subway or car. Such nodesdo not take breaks that go home. According to the BLSdata, 90% of the people drive to work, 5% walk to work,the remaining 5% take other forms of transportation in-cluding subway, bus, and other forms of transportation towork. These parameters can be altered for the simulationof European cities that have significantly lower car trips(e.g., Amsterdam 34% of the trips use cars [21]).

1The inclusion of household activ ies not p erfomed at hom e into profes-sional extends b eyond what the BLS calls receiv ing professinoal serv ices.

TABLE I. Duration at work model parameters

time α µ σ m

≤8AM 0.91 8:09 1:06 9:508-9 0.85 7:49 0:56 8:529-10 0.81 7:16 1:17 5:5210-11 1.0 7:11 2:16 -11-12 0.70 7:16 2:11 5:0012-1 1.0 6:19 2:40 -1-3 0.5 7:33 0:55 4:313-6 0.83 6:18 1:55 2:07≥6 1.0 4:30 2:26 -

Arrival time at work Figure (2) shows the comple-mentary cumulative distribution function (CCDF) of thetime of arrival at work. The observed values were fittedwith a mixture of exponential and Gaussian distribution.Specifically, with probability of 0.552, the time of arrivalis normally distributed with mean 7:46 and standard de-viation of 45 minutes. With probability (1− 0.552), thetime of arrival is exponentially distributed with the meantime of arrival of 12:00. The exponential distribution isshifted so that the earliest minimum time of arrival inthis case is 5AM. The normal distribution is truncated sothat no arrivals occur before 5AM.Duration at work Figure (3) shows the CCDF of the

duration at work for people that arrive at work between7 and 8 in the morning and for those that arrival between10 and 11 in the morning. These distributions and onesfor other arrival times at work were fitted with a mixtureof a normal random variable and an exponential randomvariable. These distributions have four parameters, α, theprobability of selecting the normal distribution, µ and σthe mean and the standard deviation of the normal distri-bution and m, the mean of the exponential distribution,which is selected if the normal distribution is not. TableI shows the value of these parameters for the differentarrival times at work. Surprisingly, while the model issimple, the fit shown in figure (3) is a typical quality fitthrough out the day. On the other hand, from Figure(2) ti can be seen that the most important distribution isthat for nodes arriving between 7 and 8.Whether a break is taken The probability of

whether a break is taken depends on the time of arrival atwork. Note that if a break is not taken, the person mayhave still eaten lunch, but they did not leave the building.We fit the probability of taking a break given the time ofarrival with a piece-wise constant probability function.

P ( taking a break| arrival time at work = t)

=

⎧⎪⎪⎪⎨⎪⎪⎪⎩0.35 for t < 6.5

0.86 (t− 6.5) + 0.35 for 6.5 ≤ t ≤ 100.17 (t− 10)− 0.65 for 10 ≤ t ≤ 130.056 (t− 13) + 0.15 for 13 ≤ t ≤ 17.5−0.08 (t− 1750) + 0.4 for t ≥ 17.5

Note that this equation is given in fraction of hours, nothours and minutes. This model and the observed proba-bility is shown in Figure (2)The time the break is started Clearly one cannot

go on a break before they arrive at work. However, oncethey arrive at work, the rate that a person goes on a break

95

TABLE II. Duration of activity model parameters

activity µ d ρ

eat 0:31 0:20 0.18shop 0:28 0:20 0.03at home 1:00 0:20 0.12professional 0:44 0:10 0.04exercise 0:35 0:20 0relax 0:27 0:15 0.01drop-off 0:19 0:10 0.02

does not depend on how long they have been at work. Fig-ure (4) shows this rate conditioned on the person arrivingat work one hour ago, two hours ago, and uncondition-ally. Only a minor difference arrises and this differenceis within the confidence intervals. Thus, we assume thatrate of going on a break is independent of arrival time,assuming that the node has already arrived at work. Therate that a person takes a break is approximated by

r (t) =⎧⎪⎪⎪⎨⎪⎪⎪⎩0.004 for t < 10.5

0.006× exp (−1.7 (12− t)) for 10.5 ≤ 120.006× exp (−0.6 (t− 12)) for 12 ≤ t ≤ 140.0058× exp (−0.3 (5− t)) for 14 ≤ t ≤ 18

0.0058 for t > 18

.

By rate of taking a break, we mean that the probabilitythat a node will take a break within the time intervalfrom t0 to t1 is (t1 − t1)

t1t0

r (τ) dτ . The observed rateand fit is shown in Figure (4).Number of activities performed during a break

Figure (5) shows the probability of performing differentnumbers of activities during a break. We see that overthe course of the day, the number of activities performedvaries. However, the variation is small, and hence we takethe probability to be independent of the time of day.Which activities are performed during a break

The types of activities performed during a break stronglydepend on the number of activities to be performed. Fig-ure (5) shows the fraction of breaks that include the in-dicated activity. Note that if more than one activity isperformed, the fractions sum to more than one.The duration of activities The time spent perform-

ing an activity depends on the type of activity. Figure (6)shows the CCDF of the duration of three activities. Thedistribution of the duration of eating shows jump at 1hour. Smaller jumps are noticeable in the distribution ofother activities. The duration of these activities and theother activities is modeled as a mixture of an exponen-tially distributed random variable conditioned on the du-ration being larger than a minimum duration along withdeterministic duration of one hour. Thus, the distribu-tion of the duration of each activity has three parameters,µ, the mean of the exponential distribution, d the mini-mum duration, and ρ, the probability of duration lastingexactly one hour. Table II shows the value of the modelparameters for the different activities considered.Once the activity has been selected, the location of

the activity must be determined. Specifically, eating re-quires selecting a restaurant, exercising requires selecting

6 8 10 12 14 16 18 20 220

0.1

0.2

0.3

0.4

0.5

0.6

4 6 8 10 12 14 16 18 20 220

0.2

0.4

0.6

0.8

1

time of arrival at work

CC

DF

prob

of ta

king

a b

reakObserved

Fitted

Figure 2. Left. Complimentary cummulative distribution func-tion (CCDF) of the time of arrival at work. Right. The proba-bility of taking a break given the arrival time at work (includingarriving after a break).

0 2 4 6 8 10 12 140

0.2

0.4

0.6

0.8

1

CC

DF

0 2 4 6 8 10 12 140

0.2

0.4

0.6

0.8

1

Duration at work (hours)

CC

DF

Arriving between 10 and 11Arriving between 7 and 8

ObservedFitted

Figure 3. The CCDF of the duration at work for two differentarrival times at work.

8:00 10:00 12:00 14:00 16:00 18:00 20:00 22:000

1

2

3

4

5

6

7

8

9 x 10-3

Time

Rat

e of

taki

ng b

reak

s (fr

actio

n of

peo

ple)

observedfixed for one hourfixed for two hoursfittedCI

Figure 4. The rate that a person takes a break and leaves workgiven the current time. Also shown are the rates conditioned onthe person being at work for at least one and two hours. Theserates are within the confidence intervals that are also shown. Fi-nally, the fitted rate is also shown.

eat

shop

at h

ome

prof

essi

onal

serv

ice

exer

cise

rela

x

drop

off0

0.1

0.2

0.3

0.4

0.5

frac

tion

of a

ctiv

ities 1

2>2

1 2 3 4 5 6 7 80

0.2

0.4

0.6

0.8

1<9am9-1010-1111-1212-11-2>2pm

number of activities`

0

number of activities

frac

tion

Figure 5. Left: the number of activities done during a break con-ditioned on the time that the break is started. Right: the fractionof time that a break includes the indicated activity given the num-ber of activities performed within the break.

96

0 0:30 1:00 1:30 2:00 2:30 3:000

0.2

0.4

0.6

0.8

1 shop

0 0:30 1:00 1:30 2:00 2:30 3:000

0.2

0.4

0.6

0.8

1eat at home

0:30 1:00 1:30 2:00 2:30 3:000

0.2

0.4

0.6

0.8

1

observedexp

0

CC

DF

duration of activity (hours)

Figure 6. CCDF of the duration of eat, shop, and at home activ-ities.

0 1000 2000 3000 400010

-2

10-1

100

meters

CC

DF

Manhattan - officeManhattan - residentialLondonChicagoSeattleEdmonton

Figure 7. CCDF of Distance Traveled During Outdoor WalkingTrips. This data is from [3].

a gym, getting professional service requires selecting anoffice location, shopping requires selecting a store, drop-ping someone off requires selecting an office location todrop them off at. It is assumed that such location typesexist, as they do in our mobility simulator as discussed inSection (III).We assume that people walk to the location that is re-

quired to perform the activity. Future work will includethe case where people take other forms of transportation.Pushkarev and Zupan [3] observed the distribution thatpedestrian walk (see Figure 7). We see that the distanceis well modeled by an exponential distribution with means554 m, 380 m, 403 m, 344 m, 813 m, and 216 m for Man-hattan from office buildings, Manhattan from residences,Chicago, Seattle, London and Edmonton respectively. Wesee that the US cities have approximately the same mean.Thus, we select a location are of the correct type (e.g.,a store for shopping) at random such that the walkingdistance is exponentially distributed with mean 400 m.

V. Task Model

Some activities consist of a single task. For example,eating consist of going to a restaurant. However, shop-ping and working consist of multiple tasks. We modelshopping as a simple random walk inside the store. How-ever, work is modeled in a more complicated manner thatfocuses on modeling meetings. Specifically, [22], [5], [23]have collect data on the frequency, size, and durations ofmeetings. [22] includes two person meetings. This workallow the model to include a large number of worker in-teractions. Thus we model mobility while at work as asequence of going to meeting followed by going back tothe persons home office. This process repeats until thework activity is complete.

TABLE III. Meetings model parameters

meeting size mean duration prob.2 21 (min) 0.653 19 0.124 57 0.045 114 0.026 37 0.047 50 0.038 150 0.019 75 0.0210 150 0.0115 30 0.02520 30 0.025

More specifically, meeting are simulated as follows. Thetime between meetings is assumed to be exponentially dis-tributed. When a meeting begins, a random number ofpeople is selected to attend the meeting. Based on thenumber of people attending, the duration of the meetingis determined. The duration is assumed to be exponen-tially distributed.The model parameters of the model are the mean time

between meetings, the distribution of the size of meetings,and the relationship between number of meeting partici-pants and the mean meeting duration. These parametersare determined from [22], [5], [23] . Specifically, the meantime between meetings is 18 minutes while Table III givesthe remaining of the model parameters.

VI. Agent Model - Node Dynamics andInteractions

Since the pioneering work of Pushkarev and Zupan [3],it has been known that pedestrians are not uniformly dis-tributed but tend to be group into clusters, or platoons.Since the distribution of nodes plays an important role inthe performance of MANETs and mesh networks, the mo-bility must also model platoons. This part of the modelis known as the agent model and is responsible for gettingthe person from one location to the next. In our simula-tor, people take the shortest path, hence path finding isnot an important part of the agent model. Rather, theagent model focuses on the dynamics and interaction be-tween moving people. More specifically, the agent modelconsist of maintaining a distance-speed relationship be-tween nodes and lane changing rules. These are discussedin the next section. In Section VI-C, the model is vali-dated by comparing the size and of platoons created bythe model to those observed by Pushkarev and Zupan.

A. Inter-node Speed-Distance Relationship

When a node with a higher desired speed catches up toa slower moving node, it will either follow or pass. To un-derstand the dynamics of catching up, it is necessary tounderstand the distance-speed relationship. The impactof this relationship is that nodes will be tightly packed(i.e. high density) if their speed is low (congestion), but ifthe speed is higher, then the nodes must be further apart(low density). Since the density of nodes plays an im-portant role in MANET performance, the distance-speedrelationship must be understood and realistically mod-eled. For vehicles, the distance-speed relationship, which

97

0 0.5 1 1.50

1

2

3

4

5

Meter/Sec

Met

ers

Mixed UrbanStudentsFitted (Mixed Urban)Fitted (Students)

Figure 8. Speed-distance relationship for pedestrians. The mixedurban pedestrian data is adapted from [24] and the student obser-vations are adapted from [25].

we denote as D (S) , is closely related to the "two-secondrule" that specifies that a following vehicle should not becloser than two second behind the vehicle it follows. Forboth vehicles and pedestrians, these relationships havebeen extensively studied. Here we focus on the pedes-trian case.The distance-speed relationship for pedestrian is

studied in [24] and [25]. Figure ?? shows thedistance-speed relationship derived from these observa-tions2 . We approximate this relationship with D (S) =S∗Dmin/ (1.08× S∗ − S) where Dmin is the minimum dis-tance between people without touching and S∗ is the de-sired speed of the pedestrian. Dmin was found to be atleast 0.35m [3].If has been found that pedestrian desired speeds are ap-

proximately Gaussian with mean 1.34 m/s and standarddeviation 0.26 [27], [28], [29].

B. Lane changing

While traffic lights are an important cause of platoon-ing3 , the passing of not passing of slower walkers alsoplays an important role [3]. A people will certainly notovertake slower walkers if there is no room (e.g., if theother lanes are full). Even if there is room, pedestrian(as well as vehicles) might not pass out of choice and se-lect to slow down and follow the node ahead [30]. Suchdecisions lead to platooning.While the dynamics of pedestrian overtaking has been

observed, it has not been modeled. However, models forvehicle passing have been developed [31]. We borrow fromthis model. It has been found that lane changing dependson the difference between the speed that results from notchanging lanes and the speed that could be achieved if alane was changed. Specifically, a slightly simplified modelfor probability of wanting to change lanes and overtake aslower node is

P (desire to change lanes) = 1/ (1 + exp (A+B (V∗ − V ∗)))

where V∗ is the speed that the node would achieve if it re-mains in the current lane and V ∗ is the speed that would

2The p lot shown is based on area-sp eed relationships w ith the assump-tion of 0 .75 meter of latera l space b etween peop le as found by Oeding[26].3Our simulator includes traffi c lights.

be achieved if the node changes lanes. Since speeds mayexperience short-term variation, instantaneous determi-nations of V∗ and V ∗ leads to erratic behavior. Instead,letting ν denote the node that is considering changinglanes, we define V∗ to be the average speed of all nodes be-tween ν and the next intersection, and V ∗ to be the min-imum of the desired speed of ν and the average speed ofthe nodes in the target lane that would be between ν andthe next intersection. Scaling the parameters found in[31], we set APedestrian = −0.225, and BPedestrian = 1.7.While this model has not been verified for pedestrians,

in the next section we will see that it does give rise torealistic platooning.

C. Validation of Pedestrian Mobility

The burstiness of pedestrians, or in the terminologytraffic engineering, pedestrian platoons, have been inves-tigated by Pushkarev and Zupan [3]. Their work hasserved as the basis for the pedestrian traffic engineeringguidelines set forth in the Highway Capacity Manual [4].The metrics of burstiness for pedestrian platoons is dif-ferent from the ones typically used in studying data net-works. Specifically, Pushkarev and Zupan compare twoflow metrics, the 15 minute average flow rate (AFR) andthe flow rate during a platoon (PFR). A node is in aplatoon if the local density of nodes exceeds the averagedensity. As is shown in Figure 9, the PFR is higher thanthe AFR. According to Pushkarev and Zupan, the largerthe PFR is as compared to the AFR, the more busrtythe pedestrian traffic. The study of Pushkarev and Zu-pan was not focused on finding the frequency of specificflow rates, but to examine what combinations of AFR andPFR occur on urban sidewalks. Thus, we use this data asa baseline with which we compare the pedestrian mobil-ity model described above. The left-hand plot in Figure 9shows two sets of data. The generated data from the mo-bility model is from a variety of configurations includingcounting pedestrians one a block with and without build-ings, various sizes of sidewalks (from 4 lanes to 32 lanes),various traffic light timings (form 60 seconds to 120 sec-ond periods), and various rates of pedestrians flowing intothe street.As can be seen from the left-hand plot in Figure 9, the

mobility model described above generates combinationsof PFR and AFR that are realistic. The center plot inFigure 9 shows the data set collected by Pushkarev andZupan and a set of data generated by the mobility modelbut where nodes pass whenever there is room to pass, i.e.,P (desire to change lanes) = 1 as oppose to what is givenin (??). Clearly, increasing the propensity to change lanesacts to decrease the burstiness so that some realistic lev-els of burstiness never occur. Finally, the right-hand plotin Figure 9 shows Pushkarev and Zupan’s data compareto data generate by the mobility model but where thereare no inter-pedestrians dynamics, i.e., nodes move alonglane irrespective of other nodes. Such mobility allows,for example, nodes to exceed the distance-speed relation-ship. As shown in Figure 9, ignoring inter-node dynamicsresults in unrealistic levels of congestion (extreme discom-fort occurs when the flow rate exceeds 7 [3]).

98

0 1 2 3 4 5 60123456789

1011

0 10 20 30 40 50 600

10

20

30

40

50

60

70

0123456789

1011

0 1 2 3 4 5 6

Observed Model

15 minute average flow rate (peds per minute per foot of width)

flo

w i

n p

lato

on

(ped

s per

min

ute

per

fo

ot

of

wid

th)

0 1 2 3 4 5 601234567891011

0 10 20 30 40 50 600

10

20

30

40

50

60

70

01234567891011

01234567891011

0 1 2 3 4 5 6

ObservedObserved ModelModel

15 minute average flow rate (peds per minute per foot of width)

flo

w i

n p

lato

on

(ped

s per

min

ute

per

fo

ot

of

wid

th)

Figure 9. 15 Minute Average Flow Rate versus Flow Rate in a Platoon. The flow rate is the number of pedestrians that pass by themeasurement point per minute divided by the width (in feet) of the sidewalk. The black line is the area of realistic values found byPushkarev and Zupan.

VII. Conclusion

The model presented here includes three sub-models,the activity model, the task model and the agent model.The activity model is based of a recent US Labor De-partment study on time use in America. The task modelmodels the mobility of people inside office buildings. Thismodel is based on work in the area of meetings analysis[5]. The agent model determines how the mobility nodeinteracts with other mobile nodes as it moves toward itsnext destination. This model is based on the work ofPushkarev and Zupan [3].While this model presents far more detail than previ-

ous mobility models, the task of modeling and simulatingurban mesh networks is far more challenging than simu-lating ad hoc networks that have been dominated by therandom way-point mobility model. This paper includesonly a model of pedestrians. While vehicles models doexists, and are included in the simulator that implementsthe mobility described here, the vehicle and pedestrianmodel are not yet integrated to the point that the modelincludes people driving and parking cars.Due to lack of space, this paper only focused on the mo-

bility model itself. Work is currently underway demon-strating the importance of such mobility models in termof network performance.

References

[1] D. Neff, “Wireless philadelphia town hallmeeting,” November 2004. available athttp://207.245.67.199/philagotit/wifilive2.wmv.

[2] S. Cass, “Viva mesh vegas,” IEEE Spectrum, 2005.[3] B. Pushkarev and J. M. Zupan, Urban Space for

Pedestrians. MIT press, 1975.[4] Transportation Research Board, 2000 Highway Ca-

pacity Manual. Washington, D.C.: National Re-search Council, 2000.

[5] N. C. Romano and J. F. Numamaker, “Meeting

analysis: Findings from research and practice,” inProceedings of Teh 34th Hawaii International Con-ference on Systems Science, 2001.

[6] A. Szalai, The Use of Time. the hague: mouton,1972.

[7] U. D. of Labor Bureau of Labor Statistics,“American time use survey (ATUS),” 2003.http://www.bls.gov/tus/.

[8] F. Bai, N. Sadagopan, and A. Helmy, “IMPOR-TANT: A framework to systematically analyze theimpact of mobility on performance of RouTing pro-tocols for adhoc NeTworks,” in Infocom, 2003.

[9] S. Bohacek and V. Sridhara, “The graph propertiesof MANETs in urban environments,” in Under Sub-mission, 2004.

[10] A. L. Cavilla, G. Baron, T. E. H. A. L. Litty, andE. de Lara, “Simplified simulation models for in-door MANET evaluation are not robust,” in Proc.SECON, 2004.

[11] T. Camp, J. Boleng, and V. Davies, “A survey of mo-bility models for ad hoc network research,” WCMC,vol. 2, no. 5, pp. 483—502, 2002.

[12] A. Jardosh, E. M. Belding-Royer, K. C. Almeroth,and S. Suri, “Towards realistic mobility models formobile ad hoc networks,” in MobiCom, 2003.

[13] J. Tian, J. Hahner, C. Becker, I. Stepanov, andK. Rothermel, “-based mobility model for mobile adhoc network simulation,” in Proceedings of the 35thAnnual Simulation Symposium, pp. 337—345, 2002.

[14] D. Batacharjee, A. Rao, C. Shah, M. Shah, andA. Helmy, “Empirical modeling of campus-widepedestrian mobility: Observation on the USC cam-pus,” in IEEE Vehicular Technology Conference(VTC), 2004.

[15] M. McNett and G. M. Voelker, “Access and mobilityof wireless PDA users,” Tech. Rep. CS2004-0780, UCSan Diego, 2004.

[16] N. Aschenbruck, M. Frank, P. Martini, and J. Tolle,

99

“Human mobility in MANET disaster area simula-tion - a realistic approach,” in 29th Annual IEEE In-ternational Conference on Local Computer Networks(LCN’04), 2004.

[17] S. Ray, “Realistic mobility for MANET simulation,”Master’s thesis, The University of British Columbia,2003.

[18] B. D. Hankin and R. A. Wright, “Passenger flowin subways,” operationsl research quarterly, vol. 9,pp. 81—88, 1958.

[19] S. P. Hoogendoorn and P. H. L. Bovy, “Pedestri-abn route-choice andb activity scheduling theory andmodels,” transportation research B, 2003.

[20] S. P. Hoogendoorn, M. Hauser, and N. Rodrigues,“Application of microscopic pedestrian flow simula-tion to station design evaluation in Lisbon train sta-tions,” in TRB 2004 Annual Meeting, 2004.

[21] C. W. Jenks, “International transit studies programreport on the first three missions,” tech. rep., TransitCooperative Research Program, 1997.

[22] R. R. Panko and S. T. Kinney, “Meeting profiles:Size, duration, and location,” in Proceedings of Teh28th Annual Hawaii International Conference onSystems Science, 1995.

[23] R. K. Mosvick and R. B. Nelson, We’ve Got to StartMeetings Like This! Glenview, Illinois: Scott, Fores-man and company, 1987.

[24] S. J. Older, “Movement of pedestrian on footwaysin shopping street,” traffic engineering and control,pp. 160—163, 1968.

[25] F. P. D. Navin and R. J. Wheeler, “Pedestrian flowcharacteristics,” traffic engineering, pp. 30—36, 1969.

[26] D. Oeding, “Traffic loads and dimensions of walk-ways and other pedestrian circulation facilities,”Strassenbau and strassenverkehrstechnik, vol. 22,1963.

[27] D. Helbing, “Sexual differences in human crowd mo-tion,” Nature, vol. 240, p. 252, 1972.

[28] D. Helbing, “The statistics of crowd fluids,” Nature,vol. 229, p. 381, 1971.

[29] G. K. Still, Crowd Dynamics. PhD thesis, universityof warwick, 2000.

[30] Y. Zhang, L. E. Owen, and J. E. Clark, “A multi-regime approach for microscopic traffic simulation,”in The 77th TRB Annual Meeting, 1998.

[31] K. I. Ahmed, Modeling Drivers’ Acceleratiuon andLane Changing Behavior. PhD thesis, MIT, 1999.

100