Impact of Buses, Taxis, Passenger Cars, and Traffic Infrastructure on Average Travel Speed

It is known that many variables inﬂuence traﬃc, yet very little is known about the weight of each factor in the dynamics of traﬃc in cities of developing countries, in many cases due to their peculiar traﬃc regulations. In this work, we search for the variables that have the most signiﬁcant impact on the average travel speed of three distinct types of vehicles: passenger cars, taxis, and buses. First, we developed a tool featuring algorithms that simulate ordinary overtaking and car-following behaviors, along with controls for setting vehicles’ actions, particularly buses’ and taxis’ stops. Then, we chose a particular zone to study, based on its common geometry and the particular traﬃc infrastructure (speed bumps, traﬃc lights, and bus stops) inside it. Later on, three experiments were carried out, with the following results. (1) Both the buses’ arrival frequency and curbside bus stops aﬀect the passenger cars’ average travel speed. The buses’ response was aﬀected by the bus bay and curbside bus stops. The buses’ speed tendency inﬂuenced neither the passenger cars’ nor buses’ response. (2) Taxis’ arrival frequency, stopping frequency, and speed tendency were found to inﬂuence the passenger cars’ response. Taxis’ response was altered by taxis’ speed tendency, while buses’ response was aﬀected by taxis’ arrival frequencies. (3) The number of speed bumps, the arrival frequency of passenger cars, and their speed conditions (homogeneous and heterogeneous) aﬀect the passenger cars’ response. We expect that the ﬁndings presented in this study, along with the recommendations made from the results, may pave the way for better road design public policies.


Introduction
e main objective of the present study is to achieve a better understanding of low travel speed. We conducted all research in an area with compromised vehicle velocity. e zone under study possesses regular traffic infrastructure, such as speed bumps, traffic lights, and bus stops. It also has regular city traffic, i.e., buses, taxis, and passenger cars (hereafter just cars). e appropriate combination of factors to sustain acceptable traffic speed (i.e., below but close to the speed limit) while avoiding undesirable traffic speed (i.e., far below the speed limit) is a functional problem under intense investigation in the field (e.g., references [1][2][3][4][5][6][7][8]). In the present investigation, we want to identify the variables that most affect the average travel speed. To do so, we will answer the following three questions: (1) Are the buses' arrival frequency and speed tendency, bus bay, and curbside bus stops the variables that influence the average travel speed of cars and buses?
(2) Are the taxis' arrival frequency, stopping frequency, and speed tendency the variables that affect the average travel speed of cars, taxis, and buses? (3) Are the number of speed bumps, cars' arrival frequency, and cars' speed conditions (homogeneous and heterogeneous) the variables influencing the cars' average travel speed?
Developing countries face serious traffic problems [9] and some of these not count with field traffic data to conduct investigations, so that new technologies as traffic simulations are a suitable (economic) choice. We employed an in-house developed simulator to model the interaction between cars, buses, and taxis. Such smart interactions, and reactions to traffic infrastructure, allowed our simulator to show which variables have stronger effects on the average travel speed. Henceforth, we performed a set of simulations to answer our research questions.
Some variables are associated with driver behavior (e.g., the speed tendency-see definition below), while others are related to traffic infrastructure (e.g., speed bumps). By detecting the variables that impact speed and how they affect it, we gain further understanding about what causes detrimental effects. Hence, we gain further understanding that may lead to better public policies. e literature review that follows presents several wellproven models that, unfortunately, fall short in answering our specific research questions, yet it proves that such a task is possible. In addition, it presents references that investigated the effect of buses, taxis, and traffic infrastructure such as speed bumps. In our work, we investigate the effect of several factors on the vehicles' average travel speed that the literature suggests have an impact on traffic. e rest of this work is organized as follows. In Section 2, we present the literature review. Section 3 deals with the method: avenue under study, design of the simulation tool, and experimental details. In Section 4, we present the statistical results of each experiment and compare the simulations' outcome (the average travel speed, also called response or output) in order to gain insight of the variables that affect it. Finally, in Section 5, we summarize the conclusions of our work and state the contributions we made.

Buses.
e implications of improper bus stop locations in New Delhi, India, were investigated in [10]. In India in 2001, the percentage of fatal crashes that occurred near a bus stop was 6.2%, and in 2009, it was 7.0% [11]. In this study, it was observed a deficiency of sufficient space to stop and perform maneuvers and also pedestrian activity outside regulations. e effect of bus stops on traffic is analyzed in [12]. is paper correlated the traffic flow stability with bus stops and normalized density. e finding in a two-lane-30 km-long highway, where vehicles cannot overtake during buses' boarding/alighting periods, is that six or more bus stops affect the stability of traffic flow. e relation between bus impact time and curb lane capacity of roadways in Beijing is presented in [13]. e bus impact time (without dwell time) is the amount of time since the bus starts to decelerate until it stops (in a bus bay stop), plus the time since accelerates to re-enter the road until it reaches another vehicle's speed. With data from 15 bus bay stops, it was found that the relation between average bus arrival frequency and average bus impact time is best explained with a power model. e work described in [14] deals with simulations of non-lane-based heterogeneous traffic. Distinct flow measurements of the vehicles' average speed for a 400 m road length (200 m on each side of a bus stop) were done. e simulation considered curbside and bus bay bus stops and dwell times of 10 s, 20 s, and 30 s. e investigation in [15] analyzes the relationship between volume and speed in the presence of bus bay or curbside stops (using midblock location) with a bus arrival frequency of 10 min. In the case of vehicles per hour per lane � 480 and curbside stops, the speed difference between dwell times (20 s, 40 s, and 60 s) is quite small (about 2.5 mph between each other). Moreover, there is no significant speed difference for the bus bay stop case regardless of the dwell time.
In [16], simulations are used to establish the impact of the bus occupancy ratio (number of buses/number of vehicles) on traffic flow capacity. It was observed that traffic flow decreases as the occupancy ratio increases. e variables studied were speed and brake response time of cars and buses, safety following-distance, deceleration capabilities, and occupancy ratio. Reference [17] also utilizes simulations that consider the space occupation rate of buses (the sum of all circulating buses' length divided by the sum of the lanes' length) and the space occupation rate of cars, in order to analyze the influence of buses on other vehicles.

Taxis.
Taxis have a significant effect on traffic, as shown in the following references. In [18], a study in Nanjing, China, identifies the percentage of taxis and cars that were involved in one of the following risk-taking behaviors: failure to yield the right-of-way, sudden lane change, inadequate stop, unnecessary passing, and failure to stay in lane. In these five situations, taxis' percentage was greater than that of cars.
In [19], the presence of taxis in a road section and its impact on vehicle dynamics was studied. Employing simulations, three taxi behaviors were explored: (1) a taxi makes a stop to pick up or drop off passengers, (2) a taxi slows down to inquire destination information, and (3) a taxi circulates at low speed seeking for customers. In another study [20], the time that taxis utilize to pick up or drop off passengers (blockage time) is investigated. Measurements were made in outlet legs of signalized intersections in Rasht City, Iran. From the observations of four intersections, the average parking maneuver was 7.37 s, and the average blockage time was 11.31 s (ranging from 8.85 s to 12.56 s).
Passenger car units per hour (PCU/h) and speed range regions were measured in the center of Patras, Greece [21]. An augmented Naïve Bayesian network was developed to associate variables. e results determine the order of importance of the relevant variables: travel speed, volume, taxis %, motorcycles%, trucks%, cars%, and buses%. It was found that the taxis percentage (among all vehicles percentages) is the major contributor to the speed-volume relation.

Bumps and Speed-Calming Devices.
In [22], the study centered on determining if speed bumps actually reduce the vehicles' speed. For this, speed observations were made in Cagliari, Italy, where the set speed limit was 50 km/h. e speed bump geometry was 30 mm high and 600 mm wide, and the speed bump was made of rubber. eir intended purpose is to protect crosswalk areas located about 20-25 m after them. In 30% of the cases, the 85 th percentile speed (at the speed bump) was higher than the speed limit, 26% lied between 45 and 50 km/h, and for the remaining cases, the speed was under 45 km/h. e social perspective towards the installation of speed bumps on Nigerian highways was studied in [23], via questionnaires. In one survey, 72% of persons answered that without bumps the vehicles' speed is very high, 14.57% responded just high, and the remaining percentages were 6.86% and 6.57%, corresponding to low and very low speeds, respectively. In another questionnaire, 52.86% answered that bump speed is low and 29.14% responded very low, while 3.14% and 14.86% corresponded to very high and high speeds, respectively.
In [24], the study centered on speed bumps and their effectiveness in reducing vehicles' speed in Nanjing, China.
e mean speed at the bump position, 500 m after, and 500 m before (control locations) was measured. T-tests showed that the speed difference at the bump and at the control locations is significant.
Reference [25] presents a study of the effectiveness of three-dimensional speed markings (i.e., painted on the road) and speed bumps. is infrastructure was intended to reduce the speed on work zones with a speed limit of 40 km/h. e speed bump effectively reduced the average speed, which was 11 km/h behind the speed bump. e average speed to traverse a 1400 m road (with speed limit � 50 km/h) before and after the installation of 3 sinusoidal speed humps and 2 chicanes was measured in [26]. Before the speed-calming measure installation, the overall mean speed was 53.5 km/h. After the installation, this value changed to 49.4 km/h, concluding that the speed-calming devices indeed have a speed-reduction effect.
In [27], modelling was done on the vehicles' speed profile in York, England.
is was modelled on roads with the following traffic calming devices: speed tables, humps, cushions, and chicanes. To explain the speed profile, the following variables were considered: entry speed, distance to the next calming device, distance from the passed device, and calming device presence (absence).

Avenue Selection.
We were interested in a place with low speed (see [28]). e selected avenue was Miguel Hidalgo, in Lerma (Mexico state), since its geometry (two lanes, same direction both), its traffic infrastructure, and the vehicle speed on it are common in Mexico state. us, the results in this work could apply to places with similar characteristics. Following [24], the avenue was divided into four segments, each of ∼ 500 m length. Specific traffic conditions occur in each segment because each has different traffic infrastructure (see Table 1). e observations mentioned in this work were extracted from video recordings made in October 2017. Additionally, the traffic lights' cycle times were set in the simulator to match the real timings closely. e speed bumps on segments S 1 , S 2 , and S 4 force the vehicles to reduce the speed to ∼10.8 km/h, a similar value as reported in [25].
Segment S 1 has two speed bumps at location 141 m (measured from the start of the segment) and 397 m.

Simulation Tool.
We choose Unity Engine (used for traffic analysis in [29,30]) to develop a simulation tool (scripts and scenario). In the simulator, variables related to vehicles can be set, such as the taxis' stopping frequency and the buses' dwell time. Also, it measures the vehicles' travel speed to traverse an avenue segment.
e C# scripts to regulate the vehicles' behavior are presented as algorithms. For overtaking, see Algorithm 1. For accelerating, decelerating, and follow another vehicle, see Algorithm 2.
e overtaking algorithm was designed to be consistent with expected human behavior. e need for overtaking is triggered when a vehicle desires to attain a higher speed than the vehicle in front, which for some reason circulates slowly, or when in the proximity there is a static obstacle, as it can be a bus or a taxi that is completely stationary. Additionally, a vehicle overtakes only when it is safe to do so. In Algorithm 1, the Overtaking() function allows a vehicle to overtake if there is no collision risk, which depends on the speed of the vehicles located ahead and behind in the other lane of the current vehicle, and if there is no vehicle at its side. e first IF detects the need to overtake comparing the speed of the current vehicle (and the speed tendency) with the speed of the vehicle in front. It also accounts for the AD of the latter. e second IF detects the convenience to overtake comparing the speed of the vehicle in front with the speed of the vehicle in front in the other lane and the distances di u and di f . e speed tendency (or preferred speed, hereafter used interchangeably) is the top speed that a vehicle will try to reach without exceeding it. e distance between vehicles is measured from the front of the current vehicle to the back of the vehicle ahead.
In Algorithm 2, ac � 2 m/s 2 for maximum acceleration and ac � −4 m/s 2 for maximum deceleration, similar to the values presented in [31]. Variable se di is defined in equation (1), br di is defined in equation (2) (an analogue formula applies to br di f ; just replace sp with sp f ), and ex di is defined in equation (3).
e constant re ti is approximately the mean perception-brake response time observed in [32]. Constant de he is in the headway mode range presented in [33], which is 1.8 to 2.2 s for densities between 20 and 80 PCU/km. If the distance between the current vehicle and the vehicle in front is less than or equal to the safe distance between the two, the current vehicle decelerates; otherwise, the algorithm applies the following rules: (1) if the headway is larger than or equal to the headway tendency plus 0.3 seconds, the current vehicle accelerates, (2) if the vehicle speed is larger than or equal to the 110% of the speed tendency, the current vehicle neither accelerates nor decelerates, (3) if the headway is less than or equal to the headway tendency minus 0.3 s, the current vehicle decelerates, and (4) if the headway is in the interval (1.8 s, 2.1 s], the current vehicle AD � 0. e speed is limited in case 2, and in case 4, a relaxation state is reached in which there is no need of taking AD actions. Clause 2 � ac f ≤ 0 � the acceleration or deceleration (AD) of the vehicle in front (ac f ) is less than or equal to zero Clause 3 � sp f < sp * 0.8 � the speed of the vehicle in front is less than the 80% of the speed of the current vehicle (sp) Clause 4 � sp f < sp u * x 1 � the speed of the vehicle in front is less than the x 1 � 70% of the speed of the vehicle in front in the other lane (sp u ) Clause 5 � di u − x 2 > di f � the distance between the current vehicle and the vehicle in front in the other lane (di u ) minus x 2 � 0 is greater than the distance between the current vehicle and the vehicle in front (di f ) ac �acceleration or deceleration of the current vehicle. de he � 2 s, headway tendency. br di �braking distance of the current vehicle. br di f �braking distance of the vehicle in front. ex di �extra distance. he �headway, the time needed by the current vehicle to reach the vehicle in front. p le � 4.2 m, car's length. re ti � 0.5 s, reaction time to brake. se di �safe distance between the current vehicle and the vehicle in front.  Journal of Advanced Transportation Even though Algorithm 1 cannot include 100% of the human reaction possibilities to decide when and how to overtake and Algorithm 2 cannot do the same for the AD actions when following another vehicle, these cover common intuitive behaviors. e vehicles considered in the simulations of the experiments that follow are of three types: cars (width � 1.7 m and length � 4.2 m), taxis (same geometry as cars), and buses (width � 2.5 m and length � 10.3 m). e variable flow for a certain vehicle type varies from one simulation to another around the observed value in the avenue under study in calm hours. us, the traffic dynamics of each simulation could be consistent with that of other avenues with equal number of lanes and comparable flow for each vehicle type. As random processes were not included in the traffic (out of the scope), there is no need to run multiple simulations per variable configuration to achieve output convergence. e simulation time of each simulation of the three experiments is 900 s.

Experiment 1: Buses' Related Variables.
Simulations in experiment 1 were configured as follows: the entrance speed of all vehicles is 40 km/h, and buses' dwell time is 30 s (the time they wait at bus stops), following the range from 8 to 35 s reported in [34]. After the dwell time, a bus in a bus bay stop tries to re-enter the circulation if possible. If not, after 10 s (by a programmed rule), a vehicle circulating in the lane adjacent to the bus stop, and behind the bus, decelerates to allow the bus to re-enter. For the curbside bus stop, the dwell time is also 30 s, during which the circulation of the vehicles behind is obstructed. Buses circulate only in the rightmost lane and therefore do not overtake. If a vehicle overtakes, it waits for 5 s to do it again in case it has to, imitating a driver who does not overtake immediately once it has; instead, it waits to understand the traffic around him before attempting to perform an overtaking maneuver once again. A car arrives per lane every 7.5 s (or per avenue each 3.75 s). e cars' speed tendency is 60 km/h. e average travel speed (ATS) is calculated with equation (4), the travel speed of the i T -th vehicle is v i T � l j /t i T , l j is the length of the j-th segment, and t i T is the time that the i T -th vehicle needs to traverse S j . Subindex T indicates the vehicle type, T � {car (c), bus (b), taxi (t)}. N T is the total number of vehicles of type T in the simulation.  Table 2, along with the average travel speed of cars (ATS c ) and buses (ATS b ). An orthogonal experimental design was adopted, following the approach in [35]. In Table 2, columns 2 through 5 conform an orthogonal array. e minimum number of simulations (NS) to obtain meaningful results is calculated with the following equation (see [36]): Index i � 1. . .k is used to enumerate cases. e i-th case corresponds with a number of levels and the number or factors with that many levels. NF i � number of factors and NL i � number of levels. In experiment 1, we have two factors with four levels and two factors with two levels, and hence NS � 9. Nevertheless, 16 simulations were performed for completeness.

Experiment 2: Taxis' Related Variables.
e simulation parameters from experiment 1 remain mostly the same for experiment 2, with the following differences: a bus arrives every 86 s, vehicles entrance speed is 30 km/h, and buses' and cars' speed tendency is fixed at 60 km/h. To perform simulations, we selected segment S 3 which has one bus bay stop and one traffic light (green time � 30 s, yellow � 3 s, and red � 27 s). e objective is to determine which variables related to taxis have an impact on the average travel speed of cars  Table 3 shows the variables' values in an orthogonal array (columns 2 to 4) and the measured average travel speed of each vehicle type. e array has three factors with three levels each; therefore, NS � 7. Nonetheless, we perform two more simulations for completion.

Experiment 3: Cars' and Traffic Infrastructure Related
Variables.
e objective is to determine the impact of (1) speed bumps, (2) the time interval between cars, and (3) cars' preferred speed (heterogeneous and homogeneous speed cases), in the cars' average travel speed. For our simulations, we selected segment S 4 , which has three speed bumps and two intersections with a traffic light in each one. Seven cars enter the avenue per intersection per traffic light cycle. e first traffic light, numbered in the traffic flow direction, has a green time � 42 s, yellow � 3 s, and red � 45 s. e second has a green time � 67 s, yellow � 3 s, and red � 20 s. e cars from upstream enter S 4 with a speed of 40 km/h and from the perpendicular streets at 15 km/h. A car decelerates to a speed of ∼ 10.8 km/h to pass a speed bump (see [25] Table 4 shows the variables' values and the output of each simulation. Columns 2 to 4 conform an orthogonal array, with NS � 6 (one factor with four levels and two factors with two levels) plus two more for completeness.

Results and Discussion
4.1. Experiment 1. ANOVA was conducted to determine the factors influencing a response (the average travel speed of a vehicle type). e analysis of variance results, with the data in Table 2, is presented in Table 5. In the following, considering 90% confidence intervals (unless otherwise mentioned), a factor with P value <0.1 influences the response, i.e., it is significant.
With ATS c as the response, factors TL b and BS1 are significant. Factor BS2 does not influence ATS c , as a bus does not interrupt the circulation when it waits in a bus bay stop. Also, as cars are able to overtake slow buses most of the time, factor SL b is not influencing the response.
If ATS b is the response, factor SL b is slightly significant (with 80% confidence intervals, 0.1866 < 0.2), while factors BS1 and BS2 (curbside and bus bay stops, respectively) both impact ATS b , as a bus waits during the dwell time either if it is in a bus bay or a curbside stop. BS2 is impacting slightly more the response than BS1, mainly because a bus spends time to re-enter in the circulation lane if it is in a bus bay stop. Factor TL b is not influencing ATS b since buses (ideally) do not interfere with each other not even at the shortest time between buses' arrival (50 s).
ATS c drops if there is a curbside stop. To measure the loss, we compare simulations in pairs, with same TL b and different BS1 (the other influencing factor). Comparing simulation 3 vs 4, the response drops to 10.74%. From 8 vs 7, it drops to 5.23%. From 12 vs 11, it drops to 3.08%. From 15 vs 16, it drops to 2.45%. en, the response's loss drops as TL b of the compared simulations increases.
Also, ATS c is influenced by TL b . We compared simulations with same BS1 � 1 and different TL b . By comparing simulation 7 vs. 4, the response drops to 6.13%. From 10 vs 7, With ATS b being the response, ANOVA shows that both BS1 and BS2 significantly affect ATS b . Comparing simulation 8 (BS1 � 0, BS2 � 0) with 4 (BS1 � 1, BS2 � 1), the response percentage loss is 61.07%. ere is no evident response difference between simulation 12 (ATS b � 21.44 km/h), with a bus bay stop (BS2 � 1), and simulation 16 (ATS b � 22.67 km/h), with a curbside stop (BS1 � 1). Considering simulations 1, 6, 15, and 12, all with BS1 � 0 and BS2 � 1, ATS b increases from simulation 1 to 12 in 20.95%, since variable SL b also increases (from 30 to 60 km/h). Table 3) are presented in Table 6. All factors impact ATS c , factor SL t affects ATS t , and factor TL t affects ATS b . In addition, considering 80% confidence intervals, TL t and TS also have an effect in ATS t , and TS also has an effect in ATS b .

Experiment 2. ANOVA results (with data in
From Table 3, simulation 4 has the lowest ATS c , with SL t and TS at minimum values. Simulation 8 presents the highest ATS c , with SL t and TL t at maximum values. us, the relevance of SL t in ATS c is made evident. e ATS c percentage gain between simulation 4 (minimum values of SL t and TS) and simulation 6 (maximum values of SL t and TS), both with TL t � 18 s, is 9.47%.
We compare simulation 2, with TL t and SL t at minimum values, and simulation 8, with TL t and SL t at maximum values, both with TS � 45 s, finding an ATS c gain of 12.86%.
We perform two more simulations: simulation 10, with TL t � 9 s, TS � 30 s, SL t � 50 km/h, resulting in ATS c � 36.7081 km/h, ATS t � 29.915, ATS b � 20.0617; simulation 11, with TL t � 27 s, TS � 60 s, SL t � 50 km/h, resulting in ATS c � 42.1927 km/h, ATS t � 36.4168, ATS b � 22.5704. Comparing simulation 10, with TL t and TS minimum values, and simulation 11, with TL t and TS maximum values, both with SL t � 50 km/h, the ATS c gain is 14.94%.
To evidence the significant impact of SL t in ATS t , we cluster simulations according to speed tendency in three groups (see Table 7). Notice that, by arranging data, first in subgroups of SL t , then inside these subgroups by ascending TS, the table self-orders in ATS t . Hence, TS is firstly affected by SL t and secondly by TS . e factor significantly influencing ATS b is TL t . In Table 3, simulations 1, 2, and 3, with TL t � 9 s, present close values of ATS b , with the average being 21.04 km/h. For simulations 4, 5, and 6, with TL t � 18 s, the response is increasing along with TS , with the average being 22.01 km/h. e same occurs for simulations 7, 8, and 9, with TL t � 27 s and average response of 22.52 km/h. en, TL t is firstly affecting ATS b , followed by TS . Comparing simulation 3 (TL t � 9 s) and 9 (TL t � 27 s), both with TS � 60 s, the response gain is 9.04%. Table 4) are shown in Table 8. All factors impact ATS c .    To compare simulations with different number of speed bumps and other factors at equal values, additional simulations were performed (see Table 9).

Experiment 3. ANOVA results (with data in
By reducing B, ATS c increases. With TL c � 6 s and SL c � HE (likely for real traffic), we compare simulations with a different number of bumps. Comparing simulation 2 (B � 3) and simulation 12 (B � 2), there is a gain of 9.95%. Comparing simulation 2 vs 6 (B � 1), the gain is 18.12%. Comparing simulation 2 vs 16 (B � 0), the gain is 30.61%. Following this comparison logic, with TL c � 4 s and SL c � HE, we obtained gains of 6.84%, 19 � 4) and ATS c � 37.74348 km/h for simulation 8 (TL c � 6). Following this idea, we calculate the response gain comparing simulation 5 vs 14 (1 speed bumps each), simulation 11 vs 4 (2 speed bumps each), and simulation 1 vs 10 (3 speed bumps each). e average percentage gain of the four cases is 2.88%. In heterogeneous traffic without speed bumps, there is a notorious response difference between a simulation with TL c � 4 s and one with TL c � 6 s, corresponding to simulation 7 (ATS c � 32.3256 km/h) and simulation 16 (ATS c � 35.0068 km/h), respectively. Following the idea, we compare simulation 13 vs 6 (1 speed bumps each), simulation 3 vs 12 (2 speed bumps each), and simulation 9 vs 2 (3 speed bumps each), to calculate the response gain. e average percentage gain of the four cases is 7.04%. en, in heterogeneous traffic, the response changes because TL c is more evident. e average speed tendency in heterogeneous traffic (SL c � HE) is 50 km/h (50% of the cars' speed tendency � 40 km/h and 50% is 60 km/h), and the average speed tendency in homogeneous traffic (SL c � HO) is also 50 km/h (100% of the cars' speed tendency � 50 km/h). Even though that in both traffic configurations (HE and HO) the average speed tendency is the same, the average travel speed is different. ere is an ATS c percentage gain of a simulation with SL c � HO over one with SL c � HE, both with TL c � 6 s and same B. Simulation 10 vs 2 (with 3 speed bumps each) presents a gain of 3.51%. Simulation 4 vs 12 (with 2 speed bumps each) presents a gain of 2.51%. Simulation 14 vs 6 (with 1 speed bumps each) presents a gain of 5.13%. Simulation 8 vs 16 (with 0 speed bumps each) presents a gain of 7.81%. e average gain of the four cases is 4.74%. Also, there is an ATS c percentage gain of a simulation with SL c � HO over one with SL c � HE, both with TL c � 4 s and equal B. Simulation 1 vs 9 (with 3 speed bumps each) presents a gain of 5.08%. Simulation 11 vs 3 (with 2 speed bumps each) presents a gain of 7.43%. Simulation 5 vs 13 (with 1 speed bumps each) presents a gain of 7.1%. Simulation 15 vs 7 (with 0 speed bumps each) presents a gain of 16.53%. e average gain of the four cases is 9.04%. e advantage of homogeneous over heterogeneous traffic is evident.

Research Questions.
e answers to our investigative questions are as follows: (1) It was found that a curbside stop and the time interval between buses' arrival (in that order of relevance) modify the cars' average travel speed. Buses' average travel speed is affected by bus bay and curbside stops.
(2) e time lapse between taxis' arrival, speed tendency, and frequency to make a stop impact (in the order presented before) the cars' average travel speed. Taxis' average travel speed is influenced mainly by taxis' speed tendency. Taxis' arrival frequency influences buses' average travel speed. (3) It was found that the three factors: speed bumps, speed tendency configuration, and time interval between cars arrival, significantly affect (in the order presented) the cars' average travel speed.

Conclusions
e ATS c percentage loss (due to the curbside bus stop) of the compared simulations with same TL b and different BS1 ranges from 2.45% to 10.74%. en, from a traffic design perspective, it is preferable to avoid curbside stops, so that buses do not make stops in a way that obstructs circulation. Instead, bus bay stops are suitable, since BS2 has no significant effect on ATS c . From the simulations compared with same BS1 � 1 and different TL b (as this variable affects ATS c ), the ATS c percentage loss is in the range 0.64% to 6.13%. erefore, TL b could be adjusted based on a supply and demand basis.
e ATS b percentage gain, if a bus travels at the minimum preferred speed (30 km/h) compared with the maximum (60 km/h), is 20.95%. us, the connection between buses' preferred speed and ATS b is evident.
As TS influences ATS c , it could be beneficial that taxis had designated spots (similar to bus bay stops for buses, or even sharing these) to not obstruct the circulation when these pick up or drop off customers. In this way, a taxi is not motivated to travel at low speed (since SL t is influencing ATS c and ATS t ) trying to find customers along the way. As TL t impacts ATS c and ATS b , if the number of taxis circulating do not match the number of customers (as it is likely in the avenue under study), it is recommended to find the TL t values (considering time and day) that balance supply and demand.
In experiment 3, we found that B affects ATS c , regardless of TL c and SL c . By gradually removing speed bumps, we noticed improvements in the response. For the avenue studied and other similar avenues, we suggest that if a speed bump is not in a space that actually benefits pedestrians, it should be removed. With B � 0 and SL c � HO, there was no response improvement when extending TL c . With B ≠ 0, for both homogeneous and heterogeneous cases, ATS c improves as cars arrive less frequently. As TL c influences ATS c and as TL c is connected with the number of cars in circulation which in turn increases year after year, persons with transportation needs could be motivated, through better public transport services and adequate infrastructure, to use other means of transportation rather than private vehicles, such as public transport or bicycles (this poses the problem of a dedicated bicycle lane). As SL c impacts ATS c and as homogeneous speeds were proven convenient than heterogeneous, it is important to have suitable traffic infrastructure (quality streets with the right capacity, proper quantity, and location of speed bumps and traffic lights, among others) since drivers partly decide their driving speed considering that, so more drivers could be motivated to reach an acceptable speed (without exceeding the legal speed limit) and keep it. e contributions are summarized as follows: (1) e simulation tool: it allows for the manipulation of the driving behavior for cars, taxis, and buses. For the last two, the frequency and period of stops can be set. e simulator offers the option to measure the speed of each vehicle separately.
(2) As the scenario to perform simulations, it was selected a common two-lane one-way avenue with regular traffic infrastructure: speed bumps, traffic lights, and bus stops. e vehicles' interaction dynamic was designed to be intuitive and implemented with algorithms. Although the variables that influence speed in other locations might not be the same than those detected in this study, the method and analysis presented in this investigation are applicable to analogue situations. (3) With the method presented, we identified among the selected variables related to each vehicle type, which most significantly affect the travel speed. (4) e results may help to identify traffic problems and to propose solutions in the realm of public policies and infrastructure improvements, and consequently, people's quality of life could be improved. Even more importantly, traffic mitigation has the added benefit of decreasing the greenhouse emissions of vehicles.

Data Availability
e simulation data used to support the findings of this study are available from the corresponding author upon request.