Isn’t This a Driving Blog?!?
Yes, dear reader, it is. The little critter seen skittering across the video poses a real problem for how we will drive in the future or, more accurately, how we won’t drive in the future. Wait, let me explain.
Teaching Cars How to Deal with Things You Don’t See Every Day
As is becoming increasingly clear, the future of driving is autonomous vehicles. Hundreds of companies and research institutes are working furiously to bring us a day where we can climb behind the wheel and head off to work while reading the paper and playing with our cell phones. Not that these activities are not already engaged in by drivers it’s just that in an autonomous vehicle, the chances of missing exits and running into things will be significantly reduced. Driverless cars have already logged millions of miles deftly avoiding other vehicles and are getting better when it comes to encounters with other types of things on the road.
With an eye toward eliminating human error accidents (the number one cause of all collisions, by the way), engineers and programmers are teaching cars how to take in information about the driving environment and then to react to them in a best case way. Detailed algorithms calculate the possibility of something happening, and the computer adjusts the car’s trajectory accordingly. The physics of potential movement scenarios of a vehicle are relatively straightforward to calculate. Other variables in the driving environment are less straightforward, but the experienced and attentive driver can react to them with ease.
Just like when you get behind the wheel, the first goal of a driverless car is to not run into things, but how do you assess possibilities and threats? If a plastic shopping bag blows across the road, you instinctively know the threat level that it poses and that even if a collision occurs, it’s not really a big deal. The problem is, how do you teach a computer how to make that same split-second judgment? All the sensors know is that something is coming toward the car traveling in an erratic pattern and now it’s little servos are whirring furiously trying to figure out how to adjust the path of the car. Think of all of the weird and unexpected things you’ve ever seen on the roadway. Can you hope to teach a computer how to deal with any of these were they to happen?
So What Does This Have To Do with the Video Again?
The detection systems used on autonomous cars are getting better at detecting “non-vehicles” in the driving environment and how to deal with them. Birds, squirrels, pedestrians and bicyclists are more recognized by the systems than ever before but, kangaroos? Not so much.
An autonomous vehicle recognizes cars and trucks as cars and trucks whether they are in front, beside or behind. To today’s detection systems, kangaroos register as entirely different objects depending on whether the animal is standing, grazing, at rest or in motion. Currently, a hopping kangaroo is perceived by these systems as being further away from the car while it is in the air than when its feet are touching the ground. Pretty hard to avoid an object when it keeps moving away from you and back again every few seconds.
The question that engineers and safety experts are asking is when does the statistical probability of something happening reach a high enough level for it to merit consideration. In other words, what is the cutoff where the chances of something happening are high enough that you need to have a plan in place to deal with it? The answer is, we’re not sure, but we’re working on it.
By the way, if after watching that video you are just glad not to be living in Australia, did you notice that the video of the free-range kangaroo was taken in Oklahoma. Ain’t life weird?
If you found yourself sharing the road recently with a car with flashing lights on top, consider erasing that memory (and ticket) with an easy online course from ApprovedCourse.com?