AVs: Some intersections may get more dangerous before they get safer

By Michael Brenneis
Many imagine a future with fleets of autonomous vehicles seamlessly traversing road networks, wirelessly connected, perfectly aware of their surroundings and other vehicles, expertly avoiding conflicts. But what happens during the transition from manually-driven to fully-autonomous cars? As the share of AVs increases, some intersections may get more dangerous before they get safer, says a new paper by Australian researchers.
The authors used microsimulation to model AV market penetration at 10 percent increments from 10 to 100 percent, in the Geelong area of Victoria, Australia. According to the model, conflicts increased at signalized and diverging diamond intersections (DDI) (figure 1) for about the first 50 percent of penetration before declining. The authors point to aggressive and dangerous behavior by manual drivers in the dilemma zone as a primary reason for conflicts. Conflicts declined consistently at priority intersections and roundabouts as the percentage of AVs increased. As AVs reached a 90 percent share in the model, conflicts at signalized intersections showed a 48 percent decrease, priority intersections a 100 percent decrease, roundabouts a 98 percent decrease, and DDIs an 81 percent decrease.

Figure 1. Conflicts between manual and autonomous cars increased during the initial ramp-up of AV share. M-M represents a manual vehicle following and interacting with another manual vehicle, A-M represents a manual vehicle following a connected and autonomous vehicle (CAV), and M-A represents CAV following a manual vehicle.

At an AV share of 10 percent the entire road network saw a conflict reduction of 56 percent (Figure 2), but most of those (84 percent) were mid-block conflicts like lane changing. Since the total number of vehicles on the road was held constant, an increased share of platooned AVs opened up more space on the road, increasing headways for manual cars. Additionally, AVs were programmed not to merge or change lanes if the result would be unsafe headways or excessive braking by other connected vehicles, and to treat yellow lights as if they were red.
AV Figure 2. Overall, conflict declines as the proportion of AVs goes up. M = manual, A = autonomous.

Initially the modeled 50 cm headway, between AVs and for AVs following manual cars, raised the potential for conflict as heterogeneity increased, but became less of an issue as AVs began to dominate the fleet. It’s possible that crash rates could have declined more slowly with more generous headways, which would have opened up less total space on the road. There is also plenty of evidence to support decreasing headways over time since decongested space on roads tends to fill up with induced demand.
The transition to a fully automated fleet is certain to be complex; but as their numbers increase, the potential that AVs have to eliminate aggressive and dangerous driving from the dilemma zone—and reduce crashes—is compelling. Of course, this analysis does not represent the full picture since it does not include pedestrians or cyclists.
Michael Brenneis is an Associate Researcher at SSTI.