100% market penetration for fully connected and automated vehicles (CAVs) will likely be post-2050, and even then may be restricted to dedicated CAV-only facilities. However, agencies need information now to make investment and policy decisions to plan for future needs.

Kittelson is leading a national pooled fund study measuring how CAVs will impact capacity at different levels of market penetration and system aggressiveness to answer how these vehicles could change our roadways as we know them. Spearheaded by the Oregon Department of Transportation and funded by 14 total DOTs who are serving as an advisory panel, the study uses microsimulation in realistic networks to predict future capacity effects on freeways, arterials, and intersections.

We recently presented these findings at the 2019 Automated Vehicles Symposium, and soon they will be available for agencies to use in planning decisions. (Results for arterial streets are expected in 2020; we look forward to sharing more!)

In a nutshell, the anticipated improvements are not as aggressive as the media sometimes reports, but our study found that automated vehicles will likely increase capacity and reduce bottlenecks. Here’s what will change.

Investigating the effects of connected and automated vehicles: Most prior studies are unrealistic, for two key reasons.

We began our study by asking what research has been done to-date. Most prior research that has been published predicts aggressive improvement to capacity (the maximum sustainable flow rate of a roadway segment, measured in throughput of vehicles per hour per lane) when CAVs are owned and operated en masse. Simulations have projected that current capacities will double, or even triple, because of future vehicles’ connectivity and quick reaction times. The media has jumped on these predictions, eager to announce how drastically freeway movement may change as CAVs or driverless cars become widespread.

While we’re certainly in favor of finding opportunities around new technologies, there are two reasons why these aggressive predictions are a bit too ambitious.

1. Study environments didn’t resemble real-world conditions.

The first problem with most CAV studies to-date is that they are researched in a controlled, isolated environment. These models are built from single-lane “pipe” facilities without on ramps and off ramps. While in this setting, capacity does improve significantly, the only place these factors impact the real world is in dedicated CAV lanes–which, while absolutely possible, are often not the bottleneck locations, making findings less important and relevant for agencies.

2. Headways are too close for comfort.

The other challenge relates to headways. When traveling at 70 mph, the current average headway (a time measurement of front bumper to front bumper of successive vehicles) of 1.5 seconds equates to about 150 feet of spacing. Studies predicting higher capacity are estimating headways of only 50 feet while traveling at freeway speeds, and that includes the width of one vehicle. That leaves us with only about 30 feet between vehicles! We (Abby and Bastian) tested an automated vehicle at the Automated Vehicles Symposium going only 15 mph. In that test, we were quite unnerved by the proximity of that following distance; we can’t imagine how terrifying that would feel at 70 mph.

Capacity is a function of perception-reaction time, physics, and level of stress. We need to consider all three components when calculating realistic headways for driverless cars.

Yes, computers can react quicker than human drivers; but automated vehicles are still at the mercy of the physics of the roadway conditions, the tire traction, and the vehicle’s brake system. We expect that human stress levels will certainly increase with closer following distances. We must account for safe and reasonable headways, recognizing there are still many factors that automated vehicles can’t change or avoid.

Will automated vehicles improve freeway traffic?

CAVs will improve freeway capacity, but at a more conservative rate.

Investigating the effects of connected and automated vehicles and their demands on highway capacity is connected to the data. To arrive at capacity predictions, we ran simulations that looked at basic two-lane vs three-lane segments, merge segments, and weaving segments with various behavioral algorithm assumptions, in order to arrange the most realistic conditions possible.

We used a federal highway algorithm based on real data (derived from actual CAVs) that incorporates lane change and formations of platoons, in which vehicles are traveling in groups and communicating with one another. (Get in touch with us if you want to talk specifics on the simulations!) Here’s what we saw for general freeway segments:

You can use these charts to trace how increases in market penetration rate (MPR) down the left-hand column improve capacity for three base capacities: 1800, 2100, and 2400 vehicles per hour per lane (vphpl).

With a base capacity of 2400 vphpl (which, today, is an ideal capacity), at 100% MPR, capacity is predicted to rise to 3200 vphpl. This is absolutely an improvement, but at a 34-36% increase, it’s a bit more realistic than the 200% or 300% increases cited in some previous studies. There is not a significant change from two lanes to three, because the number of lanes really makes very little difference to an automated driving system.

What was interesting (and affirming) is that two other presentations at the Automated Vehicles Symposium shared the same approximate capacity predictions in their research.

CAVs could significantly improve bottlenecks and reduce traffic congestion.

2400 vphpl is ideal capacity, but an important part of the study was accounting for sections of roadway that don’t have ideal capacity. This can be due to physical roadway features, such as not having a shoulder, or events that distract human drivers, like a crash on the opposite median, narrow lane widths, or anything else that may cause a driver to slow down and contribute to a lower base capacity. Interestingly, even when we started with lower base capacities in our modeling (1800 vs. 2400 vphpl), CAV data showed capacity at 100% market penetration was the same as it was for a base of 2400 vphpl.

This means the improvement for lower base capacities is more drastic. While CAV data show capacity improvement could be 34-36% for a base capacity of 2400 vphpl, improvement could be as high as 80% for a base capacity of 1800 vphpl.

Logically, the bigger improvement makes sense, because an automated driving system won’t be phased by narrow lane widths, lack of shoulder, or another distraction that would cause a human to keep larger gaps. And it’s where the biggest opportunity could be – not for road segments that are already operating at ideal capacity, but for road segments that currently experience bottlenecks and would benefit from a less curious and distracted vehicle operator.

Connectivity is an essential component of highway capacity improvement.

We also simulated how capacities would change if we took away connectivity between automated vehicles – and were not surprised to conclude this:

Connectivity is what makes capacity improvement possible.

When we took platoons out of the simulation and left every vehicle to react to its environment in isolation, capacity dropped considerably. This is because automated driving systems with no connectivity to other vehicles are more conservative than we are as human drivers, because they are programmed to maintain larger headways to address safety and liability risks. (Once you ride in an automated vehicle that’s programmed to make safe decisions, you realize how bold most human drivers are!)

Agencies will soon be able to apply our detailed CAV findings to planning decisions.

This is a big-picture summary of what we learned from simulating realistic freeway models, but there’s much more to share. Our freeway models have been presented to the pooled fund study panel and to the Transportation Research Board’s Highway Capacity and Quality of Service Committee for the purpose of updating the Highway Capacity Manual. Agencies will soon be able to apply the research to specifically plan for capacities in 2045-2050.

Advancing Highway Capacity with CAV Research

We’re excited to continue to push research in this area and support agencies in understanding the impacts of transformation technologies and how we can collaboratively plan for an efficient but safe future. Stay tuned for updates on our second phase of this research, which is looking at arterials and intersections. If you’d like to talk about the pooled fund study further, or discuss our work at the 2019 Automated Vehicles Symposium, please don’t hesitate to reach out!