Modeling Motorcycle Lane Splitting Effects on Highway Traffic
Introduction
Motorcycle movement in traffic, as well as the effects they have on traffic and the safety concerns of motorcycles in traffic are all topics that have been heavily studied and modeled over the years. In a recent study done funded by Transport & Mobility Leuven NV (TML), a Belgian transportation research firm, and done by Isaak Yperman and Kristof Carlier (2011), showed through the creation of a link-transmission model that the an increase in motorcycles in every day traffic would lead to a reduction of overall traffic congestion and emissions. Motorcycle behavior in traffic is another aspect of motorcycles that has been heavily modeled and studied. In his Doctoral thesis titled “An Agent-Based Model to Simulate Motorcycle Behaviour” done at the University of London, Dr. Tzu-Chang Lee (2007) describes the six characteristic differences in motorcycles versus passenger vehicles as being the following: Motorcycles are more likely to travel alongside another vehicle in the same lane – often other motorcycles. Motorcycles tend to follow the person in front of them in an oblique position, allowing for better visibility and safety. There is a tendency for motorcycles to tailgate to a greater degree then cars as well as pull into smaller gaps, as well as motorcycles are more likely to swerve and weave through traffic. Finally, motorcycles are more likely to use a technique known as “filtering” or “lane splitting” to move more easily through traffic – often with the end goal of reaching the head of a large queue.
Filtering and lane splitting bring me to my research topic. However first we must define these terms. The National Highway Traffic Safety Administration defines lane splitting as:
Passing between lanes of stopped or slower-moving vehicles on a motorcycle. Not permitted in most of the U.S., it is allowed in many other countries and may provide a safety benefit. Also called “lane sharing.”
The Federation of European Motorcyclists Associations further breaks down these words by defining filtering as moving between lanes in stopped traffic, while lane-splitting occurs in moving traffic. As mentioned in the NHTSA definition, lane splitting may provide a safety benefit, which feels almost counter intuitive (moving through traffic at high speeds seems like it should be dangerous). However a recent study conducted on behalf of The California Office of Traffic Safety (2014) backs up these claims through showing a decrease in motorcycle fatalities from rear-end collisions, as well as a general increase of approval of the use of this technique from other drives on the road.
This brings me to my primary research problem. While plenty of studies have been done modeling motorcycle behavior in traffic, as well as studying the risks of riding motorcycles, and in particular lane splitting, I can find little to no research on the effects of lane-splitting on traffic congestion. The purpose of this project is to try and identify whether lane splitting has a noticeable effect on traffic congestion. If yes – is there a threshold at which this effect comes in to play? Does it rely on the number of vehicles on the road, or on the ratio of motorcycles to cars on the road?
Complex systems modeling is an ideal approach for attempting to answer this question. While traffic is in itself a very complex entity with a multitude of variables, complex systems modeling allows us to strip back the confusion of such a multi-faceted system and focus on a few key characteristics of traffic flow. This model focuses on the simple parameters of acceleration, deceleration, lane preference, and ability to pass differently depending on the vehicle type. More intricate models, such as the one proposed by Shiomi et al (2013) in their paper “Modeling Mixed Traffic Flow with Motorcycles Based on Discrete Choice Approach” take on far more parameters, from discrete decision making, to placement in lane, and directional changes in lane. While a more complex model might better model the behavior, the stripped back approach used in this model allows us to better focus on the primary objective of analyzing general traffic flows.
Methods
ODD Protocol
Purpose
The purpose of this model is to simulate motorcycle lane splitting in two-lane, highway traffic with the goal of analyzing its impacts on traffic congestion.
Entities, States, Variables, and Scales
Entities – There are two forms of entities in this model: motorcycles and cars. These agents (vehicles) share a number of traits – they each have an ideal speed they would like to go (speed limit), the actual speed they are going, a degree of tolerance (patience) in which they can handle being behind a slower vehicle (or desire to pass a slower vehicle), the distance (number of patches) they can look ahead, a type identifier (0 for Car, 1 for Motorcycle) and the rates at which the decelerate and accelerate. For the purpose of this project, the rates of acceleration/deceleration, as well as the distance they look ahead are uniform for all agents regardless of type. The type identifier is decided randomly based on a percentage slider, while the degree of tolerance and speed limit are determined somewhat randomly, and finally the actual speed varies based on the running simulation. The primary difference in these agents occurs in the processes of moving through the traffic and will be discussed in further detail in the next section.
Environment – The road created for this is 101 cells long, by three cells wide, with two lanes for all vehicles, and a yellow center lane only for motorcycles. Each cell could be considered to be approximately the same length as an average vehicle. The time scale used for this is “ticks”, with one tick equating to the time it takes a vehicle to move approximately one patch at a speed of one.
Process Overview
After the initial random placements of vehicles on the road in the setup, each vehicle moves forward, attempting to accelerate to their desired speed. If they encounter a vehicle in front of them that is moving slower, they decelerate and their patience value increases. This continues until the patience value reaches the maximum patience value of the vehicle. At this point, the vehicle attempts to change lanes. If the vehicle is designated as a car, it will only change lanes if the space in the lane directly adjacent to them is unoccupied. If they vehicle is designated as a motorcycle, they will proceed into the yellow divider zone in the middle and stay there until they can find an occupied spot in their desired lane. The agents will often speed up if there is a break in the traffic, attempting to reach their desired speed limit.
Design Concepts
Basic Principles – This model is based on a relatively simple, pre-existing traffic model. It has little basis in what I have found on modeling traffic, however follows the basic behaviors of traffic in which vehicles accelerate, decelerate, and change position. This model focuses on these basic characteristics of traffic in an attempt to simply answer the stated research question. It leaves out the many complexities of a traffic system, from vehicle collisions/accidents to road conditions, in an attempt to focus on the core parameters of traffic flow to answer the research question.
Emergence – This model has a number of “built-in” results. First, and likely the most obvious of these built-in “emergent” trends, is the idea that as the number of vehicles on the road increase, the average speed of the car decreases. The other built –in trend that we expect to see from this model is the idea that as the ratio of motorcycles versus cars on the roads leans towards a larger number of motorcycles, we should see a general increase in the speed of the traffic based on the motorcycles use of the central lane.
Objectives – The only objective the agents have is to be traveling at their ideal speed. They have temporary objectives of changing lanes, but this is all to meet the end goal of traveling at their desired speed-limit.
Sensing – The agents in this model have the ability to sense if vehicles are directly in front of them, or in an adjacent lane. This plays into the decision process of deciding whether to accelerate, decelerate, or change lanes.
Stochasticity – The primary random effects in this model are found in the initial setup. The placement of the vehicles along the road, the preferred speed-limits, and the degree of tolerance are all generated randomly from a range of values. This is important in that it varies the flow of traffic – if they were not randomly assigned, the traffic would not move naturally and flow continuously like a train.
Observation – Aside from the parameter values, the data collected in the ABM is the average speed of all of the vehicles on the road at the end of the set time limit. This is collected through averaging the current speed of all of the vehicles at that point in time.
Initialization
At time zero, the world is setup with two lanes. On these lanes, agents are randomly placed. The number of agents is defined by a parameter slider, along with the ratio of motorcycles to cars. The acceleration, deceleration, and sensing abilities of these vehicles are pre-determined by parameter sliders as well. The remaining state variables of the agents are generated for each agent randomly.
Sensitivity Analysis
For this project, I ran the model changing two primary parameters – number of vehicles and ratio of cars/motorcycles. The number of vehicles ranged from 10 to 85, running at intervals of 5. The ratio went from 5% to 95% at intervals of 5%. Each parameter setup was run 10 times, leading to a total of around 2000 runs.
Evaluation Techniques
The results of the sensitivity analysis were primarily analyzed through plotting the average speed of the vehicles compared to the number of vehicles on the road using box plots to show the average results, as well as using line graphs to demonstrate to average speed compared to the ratio value at a given number of vehicles.
Results
After running the sensitivity analysis, a number of interesting results were found – both expected and unexpected. The first set of results (see fig. 1) were looking at the average speeds of vehicles based on the number of vehicles on the road. The emergent trend is a fairly simple, obvious one: as the number of vehicles on the road increases, the average speed of the vehicles generally decreases. This is true regardless of the ratio of vehicles on the road.
Fig. 1
The reasons these results make sense is relatively straight forward – think of regular traffic. As more cars are introduced to the road, there is a greater likelihood that a slow car will be in a lane. With lower volumes of traffic, the vehicles can simply move into an alternative lane and pass the slow vehicle. However as traffic volume increases, the odds that there are slow vehicles in both lanes also increases – essentially eliminating the ability for cars to move into a “faster” lane and pass, decreasing the overall speed of the traffic.
The unexpected results come in the form of the average speeds based on the ratio of vehicles on the road. From the results, we generally see little to no change in the average speeds of vehicles as the ratio of cars to motorcycles changes. In some cases, especially when the number of vehicles on the roads increases, we see a slight increase in the average speed. The below figure (Fig. 2) shows the average speeds given a particular ratio and number of vehicles on the road.
Fig. 2
The reason these results don’t necessarily make sense is fairly straightforward. One would think that if motorcycles can use the center lane as a passing lane that when there are a larger percentage of motorcycles on the roads than cars, the speed should increase because the motorcycles essentially have three lanes available, while when there are only cars on the road, they only have 2 lanes available.
Discussion
Some of the emergent trends from this model make great sense – as vehicle numbers increase, speed decreases. However, the second trend of average speed decreasing or staying relatively the same when the number of motorcycles on the road increases makes less sense. This leads me to believe there is a significant error in the methodology that was used for this model. There could be any number of factors causing this apparent trend that deviates from the expected, but if one had to hypothesis, I would guess it had to do with the method used for sampling average speed: the speed was taken as the average of all the vehicles at t = 400. However, because traffic moves through this model in a relatively wave like manor where vehicles bunch up, then release, repeating over time, I could have been sampling at random points within this wave pattern. Two possible solutions to this problem would be to either take a much larger number of runs – say 100 per parameter sweep instead of 10 – and averaging those together, or the other alternative would be to average the average speeds at every single time step over the 400 ticks.
Because of the potential flaw in the methodology, I cannot safely make any conclusions about my initial research question. If the model were working perfectly, one could state that introducing motorcycles to traffic actually increases and maintains congestion – counter to my hypothesis of a decrease in congestion. No particular threshold for this was identified through my model, but it did seem to appear from my results that as the ratio leaned towards a more car heavy road – especially when there was a higher number of vehicles on the road – that the speed seemed to increase every so slightly. Further expansion of this model, along with a much more comprehensive parameter sweep, need to be done before any concrete results can be formed though.
Conclusion
Lane splitting is a complex, controversial topic in the world of traffic analysis. Many attempts to model motorcycles effects on traffic, as well as studying the risks and rewards of lane splitting, have been taken on over the years. The goal of this modeling exercise was to try and identify whether or not lane splitting had any real effects on traffic congestion, and if there was some sort of threshold conditions for this effect to become apparent. Due to potential issues in methodology, no conclusive results pertaining to the direct question can be made without further exploration of this topic. However – we can make some conclusions based on the results of this model. First – the common sense conclusion: As traffic volume increases, traffic congestion increases as well. Second – while the results did not show the expected reduction in traffic based on an increase in motorcycle to car ratio, it did show that adjusting this ratio did have some effect on vehicle speeds. This can be seen both in the data, and visually within the model. Further testing made lead to more conclusive results in the future.
Sources
A European Agenda for Motorcycle Safety. (n.d.). Retrieved May 28, 2015, from http://www.fema-online.eu/uploads/documents/safety/EAMS2009.pdf
Glossary. (n.d.). Retrieved May 28, 2015, from http://www.nhtsa.gov/people/injury/pedbimot/motorcycle/00-NHT-212-motorcycle/glossary71-72.html
Lee, T. (2007, October 1). An Agent-Based Model to Simulate Motorcycle Behaviour. Retrieved May 28, 2015, from http://www.scribd.com/doc/25963334/An-Agent-Based-Model-to-Simulate-Motorcycle-Behaviour#scribd
MOTORCYLE LANE – SHARE STUDY AMONG CALIFORNIA MOTORCYCLISTS AND DRIVERS 201 4 AND COMPARISON TO 201 2 AND 2013 DATA. (2014, May 1). Retrieved May 28, 2015, from http://www.ots.ca.gov/pdf/Publications/2014MCLaneSplittingSurvey.pdf
Shiomi, Y., Hanamori, T., Uno, N., & Shimamoto, H. (2013). Modeling Mixed Traffic Flow with Motorcycles Based on Discrete Choice Approach. Retrieved May 28, 2015, from http://easts.info/on-line/proceedings/vol9/PDF/P318.pdf
Yperman, I. (2011). Impact analysis of an increased share of motorcycles in commuting traffic. Retrieved May 23, 2015, from http://www.tmleuven.be/project/motorcyclesandcommuting/home.htm