I Spy... Volvo's Animal Avoidance System (about Two Years Down the Road)
As an urban biologist—and a driver—I'm quite familiar with the hazards posed by the intersection of wildlife and roads. According to the Humane Society of the United States, cars are the deadliest threat to our non-human neighbors. Millions of animals, both wild and domesticated, are struck down by motor vehicles.
Even with the added protection of seat belts and a cage of steel, people don’t always walk away from the chance encounter with only a crinkled bumper and an appointment to meet with the insurance adjuster. The National Cooperative Highway Research Program estimates there are about 200 human fatalities and 29,000 injuries in the U.S. each year due to wildlife vehicle collisions (WVCs). And these figures don’t take into account the human deaths and injuries that occur when drivers swerve to avoid an animal and lose control or hit another vehicle.
Americans aren’t alone in facing this road hazard. Five minutes of research turned up some interesting numbers from around the globe:
Average Annual WVCs:
- Australia 10,000
- Canada 40,000
- Sweden 47,000
- USA 725,000 - 1,500,00
So when the folks at Car Talk Plaza forwarded a press release on Volvo’s animal avoidance system, currently in development, I wanted to know more. To my amazement, bagging an interview with Jonas Ekmark, Volvo’s Section Manager for innovative safety electronics, was as simple as dropping the Magliozzi name in an e-mail to the Swedish headquarters.
Animal-Vehicle Biologist (that’s yours truly): I understand that Volvo unveiled a pedestrian avoidance system a few years back. I also remember reading that avoiding wildlife is a bigger challenge because, while the basic size and shape of a human being is pretty standard (give or take a few inches and/or pounds), the same can't be said for non-human animals.
I assume the technology is designed to handle large animals such as deer, moose and bear, but what about medium-sized creatures—coyotes, raccoons and bobcats? Can the technology avoid smaller stuff, like a squirrel, or even something as wee as a mouse?
Jonas Ekmark: Our main objective is to save the occupants of the vehicle from injury. We focus on animals that are heavy and tall, because we know from accident research that these are frequently involved when humans get injured. This means moose, large deer, horses and similar shapes and weights. Small animals like a mouse are hard to detect. I do not expect that we will detect these with the next or even next-next generation of systems.
A-VB So what are the challenges of developing a detection system for critters with more than two legs?
Ekmark The general challenges with collision avoidance systems are: 1) to make it work reliably and efficiently when collision is about to happen, 2) to make the system stay quiet in all other situations. We spend lots of development effort on both. We create crash dummies (both human and non-human animals), we verify that these are representative to the sensing system, we "stage" different types of collisions on our test track towards the dummies. We also drive a lot in different situations, collecting all sensor data. We can re-run the data when optimizing the detection and collision prediction algorithms in our lab.
A-VB: How does the vehicle “see” the animal? Headlights do look like eyes, but I’m pretty sure that’s as far as the similarities go.
Ekmark: The sensing system is important for detection and discrimination. We currently use 77 GHz radar, 950 nm laser (invisible) and visible monochrome video connected to a powerful real-time computer vision system. We may add near infra-red-vision, and far infra-red vision (thermal image) in the near future. When combining these sources, the system has a lot of information about what's ahead of the car.
A-VB: And when the system detects an animal in the vehicle’s path… then what?
Ekmark: When the system estimates that there is a risk of collision there is a range of options. If there is still time to involve the driver (typically three seconds or more), the system can activate an audible warning or highlight the threatening object on a display. If the collision is imminent (typically less than one second ahead) the system can automatically brake hard, tighten seat belts, and prepare airbags for deployment.
Automatic braking will be used only for animals that pose a threat to the occupants of the vehicle (typically the aforementioned heavy and/or tall ones). For medium-sized animals, the system can still inform the driver with a warning. The good thing (from the non-human animal's point-of-view) is that the driver can be informed about their presence and avoid a collision.
A-VB: What about creatures who don't keep their feet on the ground—for instance, can the system handle birds?
Ekmark: Birds can be detected by the sensing system. The combined information shows whether the object has contact to the ground or not. However, we do not plan to include birds in the target set in this first generation of this technology. Large birds sitting on the ground may be detected as medium-size animals and may activate a warning.
A-VB: How does the system determine that something moving in front of the vehicle is, for example, a squirrel and not an empty plastic shopping bag?
Ekmark: The system should not do anything when it encounters debris blowing around on the road. We use different detection technologies in parallel to enhance discrimination (radar, laser, vision, near-IR, far-IR). So the combined sensor information will make the system discard the plastic bag. The good thing is that with more sensors we get not only better discrimination but also better detection and are able to handle more scenarios.
A-VB: I’m also curious about the response-time limitations. How close can the vehicle be to the animal when it appears and still successfully avoid it without causing an accident? Are we talking about a deer already standing in the road when you come around a corner, or something that darts out in front of your car?
Ekmark: The movement of an animal is always expected, measured and predicted by the system, similar to how our "pedestrian detection" system does it. However, there will be situations that cannot be handled, for instance when an animal is hidden by an occlusion until the last moment before collision. But these situations are not in majority in real life.
A-VB: How does the system predict what the animal will do in response to an oncoming car, since an unpredictable trajectory has proven helpful to prey species, evolutionarily?
Ekmark: The prediction assumes constant velocity in the ground plane, which means that the animal is assumed to keep its speed and direction until the next measurement. It is also assumed that the animal can accelerate or decelerate at a certain rate anytime. These conditions together make bounds on where the animal can be expected in the near future, even if it tries to move unpredictably.
I agree, an unpredictable trajectory as a result of evolution may be counter-productive in this case. A predator tries to hit its prey, whereas our system tries to avoid it. But the animal movements will be limited by available road friction. There are still mechanical limits to how unpredictable the animal can move. (That’s a really interesting question.)
A-VB: Hmmmm… I think there may be a secondary market among predators for the predictive component of your system--just a thought. But what kind of predications and calculations have to take place to account for other variables… the speed at which the vehicle is traveling, for instance?
Ekmark: Timing is important. We have in-depth research on collisions with elks in Sweden. The risk of severe consequences (for the vehicle occupants) is high if the elk is hit with 110 kph (70 mph) or more. But if the elk is hit with 80 kph (50 mph) or less, the outcome is very different. Then, the occupants are likely to walk away unharmed. Could the impact speed be reduced by only 30 kph (20 mph), it would make a lot a difference for the severe cases. Such a speed reduction can be achieved by one second of full braking, which would be possible in many scenarios. We know that animals are usually moving before the collision (just like pedestrians), and their trajectory will be predicted by the system. One second of braking makes a lot of difference for the injury.
A-VB: What about something that doesn't cede the road too easily, like a striped skunk? I realize you don’t have them in Sweden, but you must have a similarly confident and stubborn species.
Ekmark: The system does not rely on vision only so there is still hope for the skunk. Radar, laser and near-IR-sensors can also be used. However, we are not trying to cover 100 percent of all possible situations in real life. The paradox is that if we did, the systems would be infeasible (or much too expensive) and would thus not be used. The challenge is to be able to offer a good system on all cars. That is the way to make a difference in real life.
A-VB: Does the avoidance-equipped vehicle also monitor traffic conditions behind and to the side of the car before making an avoidance maneuver?
Ekmark: These details are not finalized yet. I agree it makes sense to do so.
A-VB: My last question is kind of tongue-in-cheek (but not completely). Could Volvo OnCall operators offer to contact a wildlife rehabilitator in the area for situations in which the car cannot help the driver entirely avoid a collision? After determining that the occupants of the car are safe and sound, of course. As a former wildlife rehabilitator myself, may I suggest this would likely be a welcome feature to add, since people usually feel badly when they hit a living creature but rarely know who to call for assistance.
Ekmark: I think it is a very good idea. I will talk to the team working with that part.
A-VB: Very enlightening—I’ll look forward to hearing that the system is ready to hit the road, so to speak. Thanks, and best of luck!