Previously I shared some movements our collared deer make across roadways. Whether you’ve been personally involved in a deer-vehicle collision (DVC) or not, I think we’ve all seen deer make stupid decisions when it comes to crossing roads. For every deer I’ve seen fleeing the roadside away from me as I approach, I’ve seen one that waits until the last possible moment before sprinting across my path. Then there is the indecisive deer that can’t make up its mind on which way to run and does an about face halfway through its escape attempt!
A 2020 study by Font and Brown coined a term that I think fits this kind of deer well: The “worst-case deer”. They wanted to test the ability of autonomous vehicles (aka self-driving cars) to detect and respond to potential DVCs. To do so, they needed to create a virtual deer that could run across their virtual road. The authors reviewed much of the same literature on deer that I did to write my last blog post – papers on deer behavior near roads, deer reactions to humans approaching on foot and in vehicles, and deer response to predators. Using the data from these studies, they created a model that would simulate the many different ways a deer could behave as it crosses a road.
They ran this model again and again with the deer making different decisions each time, e.g. how far from the road it was when it decided to cross, how fast it accelerated, if it changed direction when it perceived the vehicle, how far from the vehicle it was when it changed direction, which way did it turn, and so on. A computer program took all these simulations and found the ones that would be the most difficult for a vehicle to avoid, in other words, the “worst-case” deer.
These worst-case deer scenarios were then presented to two autonomous car controllers – one that could only brake and one that could both brake and steer – to see how often they would result in collisions. I want to remind everyone of the saying “Don’t Veer for Deer”. Swerving to avoid deer can increase the severity of injuries, so we’re going to focus on the results of the brake-only controller.
When vegetation started 4 meters from the road edge, greatly obscuring a deer’s approach, collisions became unavoidable at a speed of 33.5 mph or higher. Even at 30 mph, collisions could only be avoided 50% of the time. As blocking vegetation moved back to 10 meters from the road edge giving the car more time to detect the deer, collisions were unavoidable at 42.5 mph. And when the deer could be seen freely by the autonomous car’s sensor, which can sense objects 60 meters away without any blocking vegetation, there was a 0% chance of collisions up until a speed of 51 mph at which point there was a 50% chance of a collision.
It’s important for us human drivers to remember that these are recommended speeds for a computer that doesn’t blink, never gets distracted by kids, phones, or radios, and doesn’t need human reaction time. The takeaway is that the posted speed limit doesn’t always fit the driving conditions. The simulations in this study were based on Pennsylvania highway dimensions and speed limits often 55 mph on two-lane highways. The risk of collisions at that speed was always high. You can think of the self-driving car’s camera range of 60 meters like the headlights on your car when you’re driving at night. PSA – Don’t drive faster than you can react to a sudden obstacle at the edge of your headlights.
Fortunately, the authors of the paper point out that we’re unlikely to ever encounter the worst-case deer on an actual roadway; the worst-case deer was programmed with every POSSIBLE deer movement, not just those that are probable.
So drive safe, everyone, and may you never meet the worst-case deer!
Northern Field Crew Leader
Game Commission Deer and Elk Section