1 of 16

Lessons learned on headway management from academic studies

Dr. Kari Watkins, UC Davis

TRB Annual Meeting 2025

2 of 16

Translating Research to Practice: Implementing Real-time Control on High-Frequency Transit Routes

Work of Simon Berrebi

Full paper by Berrebi, Crudden and Watkins (2018) in Transportation Research Part A

3 of 16

Takeaways

Earlier work deriving an optimal holding policy
Berrebi, S., Watkins, K., Laval, J., 2015. A Real-Time Bus Dispatching Policy to Minimize Passenger Wait on a High Frequency Route. Transportation Research Part B: Methodological

Academic researchers often limit real-world implementations

Smaller systems
Convenient locations

Testing only in safe environment

Don’t face same scale of problems

Need to see more full-scale implementation

4 of 16

Transit Clock

Conventional safety practice incorrectly assumes failing to undertake a “correct” intended action (called a lapse) are the result of individual characteristics, such as distraction, fatigue, or impairment, rather than the result of design deficiencies. As such, they should not be concentrated at specific geographic locations, but instead distributed across the system as a function of trip generation and traffic volumes.

Under this model, crashes are regarded principally as a product of human failures (National Highway Traffic Safety Administration [NHTSA] 2015), leading to the view that crash prevention is best achieved through programs aimed at altering driver behavior, such as driver education programs and advertising campaigns that seek to discourage people from driving under the influence of drugs or alcohol or texting while driving, among others.

Design is viewed not as a means of crash prevention, but instead as a mechanism for reducing the consequences of these otherwise unpreventable human errors. Typically, this entails the application of engineering countermeasures intended to “forgive” these errors, such as wide travel lanes and roadside clear zones

5 of 16

Three Case Studies

Atlanta Streetcar

Georgia Tech

San Antonio VIA

Control Point

Conventional safety practice incorrectly assumes failing to undertake a “correct” intended action (called a lapse) are the result of individual characteristics, such as distraction, fatigue, or impairment, rather than the result of design deficiencies. As such, they should not be concentrated at specific geographic locations, but instead distributed across the system as a function of trip generation and traffic volumes.

Under this model, crashes are regarded principally as a product of human failures (National Highway Traffic Safety Administration [NHTSA] 2015), leading to the view that crash prevention is best achieved through programs aimed at altering driver behavior, such as driver education programs and advertising campaigns that seek to discourage people from driving under the influence of drugs or alcohol or texting while driving, among others.

Design is viewed not as a means of crash prevention, but instead as a mechanism for reducing the consequences of these otherwise unpreventable human errors. Typically, this entails the application of engineering countermeasures intended to “forgive” these errors, such as wide travel lanes and roadside clear zones

6 of 16

Practical Factors of Success

Conventional safety practice incorrectly assumes failing to undertake a “correct” intended action (called a lapse) are the result of individual characteristics, such as distraction, fatigue, or impairment, rather than the result of design deficiencies. As such, they should not be concentrated at specific geographic locations, but instead distributed across the system as a function of trip generation and traffic volumes.

Under this model, crashes are regarded principally as a product of human failures (National Highway Traffic Safety Administration [NHTSA] 2015), leading to the view that crash prevention is best achieved through programs aimed at altering driver behavior, such as driver education programs and advertising campaigns that seek to discourage people from driving under the influence of drugs or alcohol or texting while driving, among others.

Design is viewed not as a means of crash prevention, but instead as a mechanism for reducing the consequences of these otherwise unpreventable human errors. Typically, this entails the application of engineering countermeasures intended to “forgive” these errors, such as wide travel lanes and roadside clear zones

7 of 16

Vehicle Location Data

Vehicles sometimes incorrectly matched to opposite direction
DynamicTime functions with loss of signal for up to five minutes

Problem on the Georgia Tech Stinger Route

VIA had lag of up to 105 seconds in GPS data

Delay in displaying hold times (until system recognized vehicle had arrived) caused confusion for the operators
Edited to display based on predicted arrival times

Conventional safety practice incorrectly assumes failing to undertake a “correct” intended action (called a lapse) are the result of individual characteristics, such as distraction, fatigue, or impairment, rather than the result of design deficiencies. As such, they should not be concentrated at specific geographic locations, but instead distributed across the system as a function of trip generation and traffic volumes.

Under this model, crashes are regarded principally as a product of human failures (National Highway Traffic Safety Administration [NHTSA] 2015), leading to the view that crash prevention is best achieved through programs aimed at altering driver behavior, such as driver education programs and advertising campaigns that seek to discourage people from driving under the influence of drugs or alcohol or texting while driving, among others.

Design is viewed not as a means of crash prevention, but instead as a mechanism for reducing the consequences of these otherwise unpreventable human errors. Typically, this entails the application of engineering countermeasures intended to “forgive” these errors, such as wide travel lanes and roadside clear zones

8 of 16

Prediction

In simulation, prediction accuracy had almost no effect on either headway stability or holding time when the standard deviation of error <10%.
In all three implementations, the error was almost always within that range.

More granular data would have been necessary to implement the probabilistic version of Berrebi et al. (2015)
Control points were considered recovery time

However, drivers typically use as bathroom breaks
Problem when told to depart immediately
Holding method should add a short buffer to predicted arrival times to avoid the disruption

Conventional safety practice incorrectly assumes failing to undertake a “correct” intended action (called a lapse) are the result of individual characteristics, such as distraction, fatigue, or impairment, rather than the result of design deficiencies. As such, they should not be concentrated at specific geographic locations, but instead distributed across the system as a function of trip generation and traffic volumes.

Under this model, crashes are regarded principally as a product of human failures (National Highway Traffic Safety Administration [NHTSA] 2015), leading to the view that crash prevention is best achieved through programs aimed at altering driver behavior, such as driver education programs and advertising campaigns that seek to discourage people from driving under the influence of drugs or alcohol or texting while driving, among others.

Design is viewed not as a means of crash prevention, but instead as a mechanism for reducing the consequences of these otherwise unpreventable human errors. Typically, this entails the application of engineering countermeasures intended to “forgive” these errors, such as wide travel lanes and roadside clear zones

9 of 16

Human Element

Dispatchers informed operators of recommended holding time via radio

At control point arrival
At departure

Sometimes just missed the green phase of the signal and had to wait for an entire cycle.

When using a schedule, operators can decide to depart seconds early rather than minutes late.

Operators have a sense of operations unperceivable to holding method and even to dispatcher.

Conventional safety practice incorrectly assumes failing to undertake a “correct” intended action (called a lapse) are the result of individual characteristics, such as distraction, fatigue, or impairment, rather than the result of design deficiencies. As such, they should not be concentrated at specific geographic locations, but instead distributed across the system as a function of trip generation and traffic volumes.

Under this model, crashes are regarded principally as a product of human failures (National Highway Traffic Safety Administration [NHTSA] 2015), leading to the view that crash prevention is best achieved through programs aimed at altering driver behavior, such as driver education programs and advertising campaigns that seek to discourage people from driving under the influence of drugs or alcohol or texting while driving, among others.

Design is viewed not as a means of crash prevention, but instead as a mechanism for reducing the consequences of these otherwise unpreventable human errors. Typically, this entails the application of engineering countermeasures intended to “forgive” these errors, such as wide travel lanes and roadside clear zones

10 of 16

Route Configuration

VIA had some control points without space for vehicles to pass each other

Operator stops for a bathroom break, vehicle blocks following buses from departing, generating bus bunching from the onset

Signal priority was a hindrance

Buses > 5 minutes late typically given priority
In trial, didn’t adhere to the schedule
Signal timing introduced new element of

randomness that deteriorated quality of predictions.

11 of 16

The Role of the Driver in Headway Regularity

Work of Yerly Martinez-Estupinan

“Understanding what elements influence a bus driver to use headway regularity tools: case study of Santiago public transit system” by Martinez-Estupinan, Delgado, Munoz and Watkins (2022) in Transportmetrica A: Transport Science

“Improving the performance of headway control tools by using individual driving speed data” by Martinez-Estupinan, Delgado, Munoz and Watkins (2023) in Transportation Research Part A

12 of 16

The Driver

Fundamental element for effectively implementing headway management strategies is the driver
Bus drivers do not behave alike

Heterogeneous in how they drive and react
Willingness to comply with and implement the instructions they receive

Goals
Understand and classify bus drivers according to their influence on headway regularity
Propose operational strategies that incorporate this information to improve regularity

13 of 16

Survey of Drivers

338 drivers randomly selected from the bus company in Santiago, Chile

Causes preventing drivers from keeping headways

14 of 16

Survey of Drivers

Older and experienced drivers perceive being faster and maintaining more regular headways.
Trust their experience to know how close headway ahead is compared to the scheduled headway.
Younger drivers value having more information along the route to tell them what actions to take to reach the scheduled headway.
Instructions from devices not always able to be executed
Congestion
Bus capacity and deterioration
Heterogeneous driving characteristics across drivers

15 of 16

Driving speed to improve headway control

Driving speed heterogeneity impacts rider wait time w/ or w/o control strategy
Having accurate information about average speed of drivers significantly improves performance of headway control tools
Especially in scenarios of high variability

Allows stratification of drivers to each line according to their driving speed
Main attributes of drivers indicating average speed are age & years of experience

NC = No control, HRT = Holding with real-time

Coefficient of Variation of speed

Excess total waiting time (min)

16 of 16

THANK YOU!

Dr. Kari Watkins

kewatkins@ucdavis.edu

Associate Professor

Civil & Environmental Engineering

University of California, Davis