Simulating the Effects of Various Road Infrastructure Improvements to Vehicular Traffic in a Busy Three-road Fork
aa r X i v : . [ c s . C E ] J un Simulating the Effects ofVarious Road Infrastructure Improvements toVehicular Traffic in a Busy Three-road Fork
Marian G. Arada
Polytechnic University of thePhilippines-TaguigTaguig City 1632, MetroManila
Merly F. Tataro
Polytechnic University of thePhilippines-TaguigTaguig City 1632, MetroManila
Jaderick P. Pabico
Institute of Computer ScienceUniversity of the PhilippinesLos BañosCollege 4031, Laguna
ABSTRACT
Using microsimulations of vehicular dynamics, westudied the effects of several proposed infrastructuredevelopments to the mean travel delay time ∆ and meanspeed Σ of vehicles passing a busy three-road fork, par-ticularly in the non-signalized roundabout junctionof Lower Bicutan, Taguig City, Metro Manila. Wedesigned and implemented multi-agent-based microsim-ulation models to mimic the autonomous drivingbehavior of heterogeneous individuals and measured theeffect of various proposed infrastructure developmentson ∆ and Σ. Our aim is to find out the best infras-tructure development from among three choices beingconsidered by the local government for the purpose ofsolving the traffic problems in the area. We createdsimulation models of the current vehicular traffic sit-uation in the area using the mean travel times τ ofstatistically sampled vehicles to show that our modelcan simulate the real-world at a significance level of α = 0 .
05. Based on these models, we then simulatedthe effect of the proposed infrastructure developmentson ∆ and Σ and used these metrics as our basis ofcomparison. We found out that the proposed wideningof one fork from two lanes to three lanes has the mostimproved metrics at the same α = 0 .
05 compared tothe metrics we observed in the current situation. Underthis infrastructure development, the ∆ increases lin-early ( R = 0 .
98) at the rate of 1.03 s , while the Σdecreases linearly ( R > .
99) at the rate of 0.14 km/h per percent increase in the total vehicle volume V . Categories and Subject Descriptors
K.3.7.1 [
Computing Methodologies ]: ArtificialIntelligence—
Multi-agent systems ; K.5.1.2 [
ComputingMethodologies ]: Modeling and Simulation—
Model verification and validation ; L.9.5 [
Applied Com-puting ]: Operations Research—
Transportation
1. INTRODUCTION
The increasing number of vehicles that use the roadat the same time is causing congestions on road net-works especially at road sections where nominal widthschange such as intersections, three-way forks, androundabouts. One real-world example of such widthchange is the Bicutan Roundabout (BR) in TaguigCity, Metro Manila. Commuters passing through thisroundabout experience delay in travel everyday, espe-cially during rush hours on week days, because ofvehicular congestion in the area. BR is located at theUpper Bicutan area and has an inscribed diameter of34 m . It is bounded to the East by the Departmentof Science and Technology (DOST) campus, to theNorth by the Philippine National Construction Corpo-ration (PNCC) campus, to the West by the PhilippineNational Railways (PNR) crossing, and to the Southby North Daanghari . BR serves as the T-intersectionbetween General Paulino Santos Avenue (GPSA) whichlies along the ENE-WSW line and SLEX’s East Ser-vice Road (ESR) which lies along the NEN-SWS lineof the area (Figure 1). The nominal center of BR isapproximately 100 m WSW from DOST’s main gatealong GPSA, 150 m SWS of the ESR and San Martinde Porres fork, and 106 m ENE of PNR crossing.Jeepneys, buses, trucks, taxis, AUVs, motorcycles, tri-cycles, and bicycles are the usual type of vehicles thatpass the area. Upper Bicutan, a barangay under theurbanized City of Taguig, is situated on the westernshore of Laguna de Bay and is bordered in the southby Muntinlupa City, in the southwest by Para˜naqueCity, in the west by Cainta, Rizal, in the northeast byTaytay, Rizal and in the north by Makati City, PasigCity and the Municipality of Pateros. According to the2010 census, 62,570 out of the city’s total population of613,343 are residents of Upper Bicutan [13].The thrust of the city government administration is tomake Taguig City a “Premier City” in the Philippines.However, the physical and financial development of acity is greatly dependent, among others, on the effi-ciency of its transportation support system, particularlyhow the city manages the vehicular flow along its road igure 1: The area map of Upper Bicutan, Taguig City showing the relative location of BR [9]. networks. The efficient flow of vehicles carrying humanand non-human capitals results in the efficient flow ofresources. This translates to a productive citizenry andsuccessful businesses. Because of this, the city govern-ment must make its move in pursuing efficient and prac-tical steps to at least mitigate, if not totally eradicate,traffic congestion which must result into improved vehic-ular flow. Vehicular flow may be quantified by the fol-lowing metrics and we say that improvement has beenachieved if there is a statistically significant differencethe measurement at a particular significance level α ascompared to the current situation:1. Average vehicular delay time while in BR (∆) –This metric measures the total amount of timethat the vehicle is in the stopped position (i.e.,when the vehicle has a speed of 0 km/h ); and2. Average vehicular speed while in BR (Σ) – Thismetric measures the mean speed of a vehicle whiletraveling within the BR area.Ideally, we wanted ∆ to approximate zero while wewanted Σ to be near but not more than the legal max-imum speed. In realistic situations, however, ∆ > > > make its move :(1) Improve the infrastructure of the road networkthat will best optimize the ∆ and the Σ, or (2) imple-ment responsive but efficient time-dependent trafficschemes that will significantly improve both metrics.Both moves are very costly if the government is totry all possible schemes and infrastructure just for thesimple reason of finding the optimal one.In terms of finding the optimal infrastructure improve-ment for BR, one should not do it via a trial-and-errorway because it means the following: (1) building theinfrastructure, (2) letting it tried by the commuters fora reasonable amount of time just to statistically compilethe ∆ and the Σ metrics, and then (3) removing it tobuild the next one. By the time the government hascompiled the data it needed, it has already spent sig-ificant amount of money, time, and other resources.Besides, the data collected might have already beenbiased due to what we call “data collection dependency”brought about by the memory of the agents in theirexperiences. This means that for a trial-and-error with n iterations, the data collected on the ( i + 1)th iteratemight have been interactively affected by the improve-ment made during the i th iterate, unless of course, afterevery iterate we revert back to the current infrastruc-ture for the same amount of time just so both the driversand the commuters will forget their i th experiences.In terms of implementing responsive yet efficient trafficschemes, most local governments have the habitualliking of doing it in the trial-and-error way, as expe-rienced by all of this paper’s authors. If for examplethe local Traffic Management Office (TMO) is consid-ering n traffic schemes, they will: (1) implement the i th scheme for a t period of time (where t is empir-ically selected to be long enough for the drivers andthe commuters to adapt to it), (2) compile the ∆and the Σ metrics, and (3) repeat the process for the( i + 1)th scheme. The local TMO will then comparethe schemes and decide based on the metrics collectedwithout considering (maybe even knowing) that thereis an interactive effect on the ( i + 1)th scheme theexperiences and adaptation of the driver and commuteragents during the implementation of the i th scheme.One efficient way that may be used to try out ideasand schemes in solving traffic congestions is to usecomputer-based simulation, a model created to reflectreal-world situations. Once the model is constructed,solutions can be tried out to determine its impact ontraffic congestion [3], without the need for a costlyreal-world trial-and-error implementation. In thisstudy, we used a multi-agent framework to simulate theobject-following behavior of heterogeneous agents thatshare a road network. The road network reflects thescaled-down dimension of a real-world network, suchas that of the BR, servicing scaled-down dimensionsof real-world vehicles. Our model can be character-ized as microscopic, stochastic, discrete time-step, andbehavior-based. The model is microscopic because itsimulates the behavior of the vehicles themselves whileon the road in the presence of, and interacting with,other vehicles. The model is stochastic because it isbased on random events that follow some distribu-tions that we observed in the real-world. The modelis discrete time-step because at each unit time, theagents concurrently change their respective positions,directions, and speeds driven only by their respectivesimulated behaviors in response to their respectiveinteractions with the current state of the road network.The model is behavior-based because the traffic flowalgorithms that we used are based on a psycho-physicalcar following model for vehicles moving along the lengthof the road [6, 8]. We used a rule-based algorithm forlateral movement that simulates the lane-changingbehaviors of the agents [1, 5].We present in this paper the result of our experimentsin searching for the optimal infrastructure improvementfor BR from among three choices that are being consid- ered by the local government. We present our results insearching for the optimal traffic schemes for the currentinfrastructure in another paper [12]. Our study hadthree objectives: (1) To show that our model reflectsthe current real-world scenario at a statistical signifi-cance of α = 0 .
05 using the mean travel time τ of thesampled vehicles as metric, (2) to use the model to sim-ulate the effect on ∆ and on Σ of three infrastructuredevelopments (ID) in BR currently being considered bythe local government, and (3) simulate the effect on ∆and on Σ the 10%, 50%, and 100% increase in vehic-ular volume under the best ID. Based on our results,we found out that: (1) The average travel time of thesimulated vehicles while at the BR is not significantlydifferent at α = 0 .
05 from the observed average traveltime of the sampled real-world vehicles, (2) the bestID is when the South bound lane of the ESR is widenup to three lanes continuing to the west bound lane ofGPSA from BR to PNR, and (3) the ∆ increases linearly( R = 0 .
98) at the rate of 1.03 s per percent increasein the total vehicle volume V , while the Σ decreases lin-early ( R > .
99) at the rate of 0.14 km/h per percentincrease in V .
2. MODEL
A computer model that controls an agent’s behavior(i.e., the driver) in terms of its reaction to the vehicle infront of it while at the same lane is called a car-followingmodel [10]. An agent-driven vehicle is said to be fol-lowing when it is constrained by the speed of the movingvehicle in front of it, and that driving at the agent’sdesired speed will lead to a collision. When a driveragent is not constrained by another vehicle, it is saidto be free and travels, in general, at its desired speed.The actions of the following agent is defined by theagent’s acceleration, although some models, for examplethe car-following model developed by Gipps [8], definethe agent’s actions through the agent’s speed. Somecar-following models only describe the agents’ behaviorwhen they are currently following another vehicle, whileother models determine the agents’ behavior in all situa-tions. We believe, however, that to model the real-worlddecision-making capabilities of human drivers, a car-following model should identify both of the following:1. The current state S the vehicle is in; and2. What actions A are desirable at this state.Most car-following models use several states to describethe following agents’ behavior. Most models use:1. S f : A state for free driving , where the vehi-cles are unconstrained and the respective driveragents try to achieve their desired speeds (subjectof course to pertinent legal speed limits of the roadnetwork);2. S n : A state for normal following , where the fol-lowing agents adjust their respective speeds withrespect to the speeds of the vehicles in front ofthem; and. S e : A state for an emergency deceleration ,where agents try to decelerate to avoid a collisionwith the vehicle in front.Throughout the paper, we used the notations summa-rized in Table 1 to describe the kinematic quantities andmodel outputs. S f Free Driving State
The agents at the S f state try to accelerate or decel-erate to achieve its current desired speed. If the currentspeed v i of the i th vehicle is higher than its desiredspeed v desired i , the agents uses the normal decelerationrate ( − a normal i ) to slow down to the desired speed. If v i < v desired i , the agents use its maximum accelerationrate a max i to reach v desired i at the shortest time possible.The − a normal i and a max i are parameters of the simula-tion and they are dependent on the type of vehicle theagent is driving. The acceleration rate of the i th agentis expressed as a i = a max i v i < v desired i v i = v desired i a normal i v i > v desired i The time t i it will take for the i th vehicle to achieve itsdesired speed is t i = ( v i a max i v i < v desired iv i a normal i v i > v desired i S n Normal Following State
While at the normal following state S n , the accelera-tion rate of the i th vehicle is given by an asymmetricalGazis-Herman-Rothery (GHR) model [4, 7, 14]. Theacceleration is computed as a i = r ± v s ± i ( x i − − L i − − x i ) t ± ( v i − − v i ) , where r ± , s ± , and t ± are model parameters. Theparameters r + , s + , and t + are used if v i ≤ v i − , whilethe parameters r − , s − , and t − are used if v i > v i − .Notice here that the final acceleration of the i th vehicleis given as max { a i , a i − } . S e Emergency State
Under this state, the agents use a deceleration rate thatprevents collision and extends δx . This deceleration rateis given by a i = ( min { a normal i , a i − − . v i − v i − ) x i − − L i − − x i } v i > v i − min { a normal i , a i − + 0 . a normal i v i ≤ v i −
3. METHODOLOGY
We present in this section the activities that we per-formed in this study. We also present here the metricsthat we measured, as well as the statistical analysis thatwe employed to analyze the results of the study.
We conducted interviews in order for us to understandthe current plans of the two local governments that havejurisdictions over the BR: the Taguig City Traffic Man-agement Office (TMO) and the Metro Manila Develop-ment Authority (MMDA). The interview served as somesort of leveling off with the intended stakeholders. Wealso wanted the result of this research endeavour to beput to use, particularly because two of the three authorsof this paper pass this area on a daily basis. Thus,if either or both local governments will implement theresult of our research, we will greatly benefit from such.We also asked permission from the two governments toallow us to conduct scientific observations so that wecan quantify the current vehicular flow in BR.
We employed the help of twelve enumerators whom weassigned to different identified points within the BRjunction. Each point of entrance to and exit from BRhas designated persons to record the time of entranceand exit of the vehicle, and the type of vehicle thatpassed through. Each vehicle is identified by their tagor plate number. We recorded a series of actual obser-vations during peak and non-peak periods for us to beable to get a statistically accurate data and determinethe highest and lowest traffic volume. From these obser-vations, we were able to obtain the mean travel time τ ofsampled vehicles, as well as the respective distributionsof each vehicle type. To identify the specific points inthe BR, we divided the area into six routes as follows:1. Route 1 is the route from PNR to DOST CampusEastbound along the GPSA passing through theBR;2.
Route 2 is the route from PNR to PNCCCampus, Eastbound along GPSA from the PNRto BR, and then turning left through the BR, andup to the Northbound lane of ESR going towardsthe PNCC Campus;3.
Route 3 is the route from DOST campus to PNRWestbound along the GPSA passing through theBR;4.
Route 4 is the route from the DOST Campus tothe PNR Crossing, Westbound along the GPSAup to the BR, and then turning right through theBR, and up to the Northbound lane of ESR goingtowards the PNCC Campus;5.
Route 5 is the route from PNCC Campus toPNR, Southbound along the ESR, turning rightthrough BR, and then Westbound along theGPSA to the PNR; and6.
Route 6 is the route from PNCC Campus tothe DOST Campus, Southbound along the ESR,turning left through BR, and then Eastboundalong the GPSA to the DOST Campus. able 1: Alphabetical listing of notations, mathematical variables, and abbreviations used in thispaper including their respective descriptions.
Notation Description δx Space headway between x n − and x n in mδv difference in speed between x n − and x n in m/sa n Acceleration of the n th vehicle in m/s L n length of the n th vehicle in ms n Effective length of the n th vehicle in mT Reaction time in sv n Speed of the n th vehicle in m/sv desired n Desired speed of the n th vehicle in m/sx n Longitudinal position of the n th vehicle in mα Statistical level of significance∆ Mean delay time per vehicle in s Σ Mean vehicular speed in m/sτ
Mean vehicular travel time sτ o τ of the sampled vehicles from observation τ s Mean τ of the simulated vehicles V Total volume of vehiclesANOVA Analysis of Variance using the F-StatisticsBR Bicutan RoundaboutDOST Department of Science and TechnologyESR East Service RoadGPSA General Paulino Santos AvenueID Infrastructure DevelopmentMMDA Metro Manila Development AuthorityPNCC Philippine National Construction CorporationPNR Philippine National RailwaysPUP-T Polytechnic University of the Philippines-TaguigSLEX Southern Luzon ExpresswayTUP-T Technological University of the Philippines-TaguigTMO Traffic Management Office
We conducted a replicated microsimulation study of thevehicular flow under the current BR using the data onrespective distribution of vehicles by type. We repli-cated the study to n = 10 and computed the mean τ .We wanted to know if the model can statistically repro-duce the observed vehicular traffic data. Statistically,the respective differences of the τ between the observed( τ o ) and the mean simulated ( τ s ) runs must not be dif-ferent from zero at α = 0 .
05. We used analysis of vari-ance statistics (ANOVA) to evaluate two hypotheses,the null hypothesis H and the alternative hypothesis H a as follows: H : There is no significant difference between τ o and τ s at α = 0 . H a : There is significant difference between τ o and τ s atthe same α . We created the respective replicated n = 10 microsim-ulation studies of the vehicular flow using the observedcurrent vehicular distribution when under the differentplanned infrastructure developments (ID) of the BR.These IDs are based on the C6 Project blueprint, themedium-term development plan by the City Mayor of Taguig Lani Cayetano, and DOST’s Monorail Projectwhich is currently underway. The proposed develop-ments are:1. Infrastructure Development 1 (ID1): Widening ofthe GPSA into three lanes but only those lanestowards the BR. That is the east bound lane fromPNR to BR and the west bound lane from DOSTto BR.2. Infrastructure Development 2 (ID2): Wideningof the GPSA into three lanes but only the east-bound lane from PNR to BR continuing from BRto DOST gate.3. Infrastructure Development 3 (ID3): Widening ofthe south bound lane of the ESR into three lanesup to the BR continuing to the west bound laneof GPSA from BR to PNR.We computed the respective means of ∆ and Σ for allinfrastructure improvements, including that of the cur-rent infrastructure we termed here as ID0 (i.e., fromID0 through ID3) and conducted an analysis of vari-ance (ANOVA) statistics to find out whether the respec-tive means are significantly different from each other at α = 0 .
05. That is, we want to test two hypotheses, thenull hypothesis H and the alternative hypothesis H a asfollows: : The absolute pairwise differences | ∆ ID0 − ∆ ID1 | = · · · = | ∆ ID1 − ∆ ID2 | = | ∆ ID1 − ∆ ID3 | = | ∆ ID2 − ∆ ID3 | =0, where the = sign in this sentence would mean “notsignificantly different from” or “statistically equal to.” H a : Any of the following respective pairwise absolutedifferences is true: | ∆ ID0 − ∆ ID1 | > . . . , | ∆ ID1 − ∆ ID2 | > | ∆ ID1 − ∆ ID3 | >
0, or | ∆ ID2 − ∆ ID3 | > We rerun the respective replicated microsimulationstudies under ID0, ID1, ID2, and ID3 but with respec-tive increases of 10%, 50%, and 100% in V , making surethat the distribution by vehicular type is preserved.Here, we wanted to know if the benefits of the proposedimprovements will be carried with the increase in V andif so, up to how much increase. The assumed increasein V is just a natural reaction of the driving agentswhen there is a perceive improvement in the currentsituation. The improvement of the vehicular flow due toinfrastructure improvements will attract more agents,increasing V in the area, and thereby might degradethe expected designed benefits of the improvement inthe long run.
4. RESULTS AND DISCUSSION4.1 Actual Current Vehicular Flow
Tables 2 and 3 summarize the distribution of vehicleper route and per vehicle type, respectively, that wecompiled during the actual observation. Route 4 hasthe most number of total vehicles that passed duringthe observation period with 23.79%. This is followedby Routes 5, 1, and 2, in non-increasing order. Route4, undeniably has the most number of vehicles sinceRoute 4 is the only route that may be used by residentsfrom the Lower Bicutan area who are trying to go tothe Metropolitan Manila Area via the C-5 Diversion.Aside from the Lower Bicutan residents, the route isalso extensively used by students and employees of sev-eral Higher Education Institutions (HEIs) and govern-ment agencies such as the Technological University ofthe Philippines-Taguig (TUP-T), the Polytechnic Uni-versity of the Philippines-Taguig (PUP-T), the LGU-funded Taguig City University, and DOST’s nationaloffices and laboratories. Route 5 is extensively usedby trucking companies with medium-length trucks car-rying heavy cargoes. These vehicles are coming off fromthe C-5 Diversion going to the Southern Luzon Areavia the SLEX. Route 1 is obviously used by commuterscoming from the SLEX area going to the residentialLower Bicutan or to any of the HEIs and governmentagencies. Route 2 is the reverse of Route 5, wheremostly medium-length trucks pass through carrying car-goes from the Southern Luzon area via SLEX going toMetropolitan Manila Area through the C-5 Diversion.From Table 3, motorcycles and cars combined accountfor about 70% of the vehicles that use BR. This is fol-lowed by bicycles, jeepneys and vans. The heavier vehi-cles, such as the many-wheelers and buses only accountfor about 4% combined. This distribution by type isactually expected because the area is used extensively by residents and commuters rather than by non-humancargoes. Table 4 shows the mean vehicular dimensionsby vehicle type. We used these dimensions to scale therespective agents in our simulations.
Table 2: Distribution of the total number ofvehicles that passed through the BR during theobservation period.
Route Number of Vehicles Percentage (%)
Route 1
778 20
Route 2
719 18
Route 3
524 13
Route 4
934 24
Route 5
815 21
Route 6
156 4
Total
Table 3: Distribution of the vehicles per type.
Type Percentage (%)Motorcycle 38.304 × × Table 4: The average dimension of observedvehicles per type.
Vehicle Length WidthType ( m ) ( m )Motorcycle 2.00 1.54 × × Table 5 summarizes the statistics we computed afterconducting ANOVA on τ o and τ s . Since we see that theF statistics α F = 0 . > α = 0 .
05, then we acceptthe null hypothesis H and we say that the differencebetween the mean sampled observed data τ o and themean simulated data τ s is not significantly different fromzero at α ≈ .
05. Thus, our microsimulation model wasable to mimic the driver’s behavior in the real worldwith a guarantee of being correct 1 − α = 0 .
95 of thetime.
Tables 6 and 7 show the respective ANOVA tables ofthe mean ∆ and mean Σ for ID0, ID1, ID2, and ID3where we find that α ∆F < α and α ΣF < α . Based on the able 5: The ANOVA table comparing τ o and τ s using the F statistics. SOV means Source of Variationand DF means Degree of Freedom SOV DF Sum of Squares Mean Square F α F Replication 9 1,633.05 181.45 39.21 < τ o vs. τ s α ∆F and α ΣF , we accept the alternate hypoth-esis H a saying with confidence 1 − α F = 0 . ID x and ∆ ID y , and between Σ ID x and Σ ID y , ∀ x = y , issignificantly greater than zero. Figures 2 and 3 show therespective mean ∆ and mean Σ of ID x , ∀ x = 0 , . . . , Figure 2: The respective mean ∆ of the var-ious IDs with DMRT grouping. Means with thesame letter are not statistically different fromeach other by DMRT. Figures 4 and 5 show the respective effect of increasing V from 10% to 100% on ∆ and Σ while under ID3. As V is increased one percent at a time, ∆ increases linearlywith R = 0 .
98 at a rate of 1.03 s while Σ decreaseslinearly with R > .
99 at a rate of 0.15 km/h . At thislinear rate of decrease, all vehicles will come to a halt(i.e., Σ = 0) at
V ≈
5. CONCLUSION
In this paper, we presented a multi-agent-basedmicrosimulation of a real-world vehicular traffic basedon the existing psycho-physical car following model
Figure 3: The respective mean Σ of the var-ious IDs with DMRT grouping. Means with thesame letter are not statistically different fromeach other by DMRT. for simulating the movement of vehicles along thelane [6, 8] and a rule-based algorithm for simulatingthe lane-changing behaviors of the vehicles [1, 5].We simulated a real-world infrastructure involving sixroutes merging into a three-road fork via an unsignal-ized roundabout. This roundabout is the BR located inTaguig City, Metro Manila. From the observed distri-bution of vehicles per route and per vehicular type, wesimulated the vehicular traffic of the current situationand collected the τ metric. Based on the ANOVA, wefound out that τ o and τ s are not statistically differentfrom each other based on the F statistics. This meansthat our simulation model can represent the real-world.We then simulated three proposed infrastructure devel-opments (ID1, ID2, and ID3) and found out that ID3offers the least ∆ and at the same time the fastest Σ.We increased the V from 10% to 100% under ID3 andfound out that ∆ increases linearly with V , while Σdecreases linearly. Based on these results, we providethe following conclusions:1. Our microsimulation model can stastically repre-sent the real-world;2. The best ID for BR is ID3; and3. Under ID3, V has a positive linear effect on ∆ anda negative linear effect on Σ.
6. RECOMMENDATIONS
We recommend to the stakeholders within BR, suchas the City Government of Taguig through its TMO,as well as the Engineering Section of the MMDA, theresidents of Lower Bicutan, the public and private able 6: The ANOVA table comparing the different IDs in terms of ∆ using the F statistics. SOVmeans Source of Variation and DF means Degree of Freedom SOV DF Sum of Squares Mean Square F α ∆F Replication 9 2,328.63 258.74 5.12 0.0004ID 3 3,556.14 1,185.38 23.47 < Table 7: The ANOVA table comparing the different IDs in terms of Σ using the F statistics. SOVmeans Source of Variation and DF means Degree of Freedom SOV DF Sum of Squares Mean Square F α ΣF Replication 9 14.73 1.64 5.63 0.0002ID 3 11.83 3.94 13.56 < Figure 4: The respective mean ∆ of ID3 when V was increased from 10% to 100%. The line rep-resents the regression with ∆ = 1 . × V + 72 . ( R = 0 . ). employees within BR such as the DOST, the TUP-T,and the PUP-T, and the private sector, particularlySM Corporation, to help call for the implementationof infrastructure developments described in ID3. Wecaution, however, that we may be able to find a bettervehicular flow if hybrid improvements of ID1 throughID3 are considered. We further caution that it maybe cheaper to implement signalized traffic schemestogether with the ID. As we have mentioned before,we have already reported our findings with varioustraffic schemes elsewhere [12], while we are currentlyworking [2] on the interactive effects of ID and trafficschemes to both ∆ and Σ.
7. FUTURE WORK
Several scientific research efforts have already forked outfrom this initial investigation. These are:1. Studying the effects of installing railings in busysidewalks;2. Pedestrian dynamics for a very, very large aca-demic building beyond its designed full capacity;
Figure 5: The respective mean Σ of ID3 when V was increased from 10% to 100%. The line rep-resents the regression with Σ = − . × V + 26 . ( R > . ).
3. Evaluation of evacuation plans for Tsunami-vulnerable shoreline communities;4. Designing re-routable and adaptive traffic plansfor time-responsive traffic schemes; and5. Incorporating the “greedy” behavior of Jeepneyand tricycle drivers in the car-following model.
8. ACKNOWLEDGMENT
This research effort is funded by the DOST-SEI Grad-uate Scholarship Program with M.G. Arada andM.F. Tataro as scholar-grantees taking the Masterof Information Technology graduate program in theUniversity of the Philippines Los Ba˜nos under theresearch supervision of J.P. Pabico. We thank the PTVGroup [11] for allowing us to use their visual simulationframework free of charge for academic and researchpurposes.The following are the respective contributions of theauthors:. M.G. Arada implemented the computational solu-tion, conducted the real-world observation, thecomputational experiments and the statisticalanalyses, and prepared the final manuscript;2. M.F. Tataro implemented the computational solu-tion and conducted the real-world observation andcomputational experiments; and3. J.P. Pabico formulated the computational solutionto the problem, conducted and interpreted the sta-tistical analyses, and edited the final manuscript.All authors declare no conflict of interest.
References [1] Talal Al-Shihabi and Ron R. Mourant. Towardmore realistic driving behavior models forautonomous vehicles in driving simulators.
Trans-portation Research Record , 1843:41–49, 2003.[2] Marian G. Arada, Merly F. Tataro, and Jaderick P.Pabico. Simulating the interactive effects of infras-tructure improvements and traffic schemes to vehic-ular travel time in the bicutan roundabout.
Trans-actions of the National Academy of Science andTechnology, Philippines , 36(1), 2014. (accepted).[3] Joe G. Bared and Evangelos I. Kaisar. Medianu-turn designs as an alternative treatment for leftturns at signalized intersections.
Institute of Trans-portation Engineers Journal , 72, 2002.[4] Robert E. Chandler, Robert Herman, andElliott W. Montroll. Traffic dynamics: Studies incar following.
Operations Research , 6(2):165–184,1958.[5] Martin Fellendorf and Peter Vortisch. Microscopictraffic flow simulator VISSIM. In Jaume Barcelo,editor,
Fundamentals of Traffic Simulation , pages63–94. Springer, New York, 2010.[6] Hans-Thomas Fritzsche. A model for traffic sim-ulation.
Transportation Engineering Contribution ,5:317–321, 1994.[7] Denos C. Gazis, Robert Herman, and Richard W.Rothery. Nonlinear follow the leader models oftraffic flow.
Operations Research , 9(545–567), 1961.[8] Peter G. Gipps. A behavioural car-following modelfor computer simulation.
Transportation ResearchPart B: Methodological
Proceedings of the 2nd International Sym-posium on the Theory of Road Traffic Flow , pages1–13, 1963.[11] PTV Group. The Official Website of PTV Group,2013. http://company.ptvgroup.com/en. [12] Merly F. Tataro, Marian G. Arada, and Jaderick P.Pabico. Selecting the best traffic scheme for thebicutan roundabout using microsimulations of mul-tiple driver-agents. In
Proceedings of the 6th UPLBCAS Student-Faculty Research Conference , NCASAuditorium, UPLB, 2013. CAS Standing Com-mitte on Research.[13] The City Government of Taguig. The Official Web-site of Taguig City, Metro Manila, Philippines,2013.[14] Qi Yang and Haris N. Koutsopoulos. A microscopictraffic simulator for evaluation of dynamic trafficmanagement systems.