Researchers have taken the next step toward the self-driving car with new image-recognition tech.
When I was a kid I was promised a flying car. When I became an adult that promise became a bit more sensible and changed ‘flying’ into ‘self-driving,’ and yet I am left with neither.
So, as far as I’m concerned, science owes me one. All of us, really.
We all deserve to catch up on our reading on the way to work. Who really wants to be a designated driver?
Despite what we deserve, scientists have encountered some pretty big roadblocks on the way to the autonomous car.
The number of quick judgment calls that we make on the road may seem routine to many of us veteran drivers, but they are still amazing enough that it is very difficult to test an AI in the same situation.
Those quick judgment calls require extremely fast data recollection, which, for a computer, is comprised of storage as well as active memory.
If we could just improve the speed with which AI can take in an image, access their memory for a comparison, analyze the data, and come up with a judgment, we would be able to make the robo-chauffeur that has the stuff to become the holy grail of the automobile industry.
Thankfully, for the part of me that wants to believe all of my childhood promises, researchers may have found a solution with a special kind of electrical resistor called a Memristor.
A Good use for Specialized Memristor Tech
The memristor is being developed by researchers at the University of Michigan using an idea taken straight from nature’s playbook known as sparse coding.
Sparse coding is a process that mammals use for vision where different shapes are organized by their characteristics, neurologically speaking.
According to lead researcher Wei Lu, “A mammal’s brain is able to generate sweeping, split-second impressions of what the eyes take in.
One reason is that they can quickly recognize different arrangements of shapes. Humans do this using only a limited number of neurons that become active.”
The memristor mimics this ability by first being trained to accept certain images as templates so that similar visual cues will allow it to recognize something that could vary widely in nature but still be the same thing, like people, animals, and hopefully in the future, other vehicles.Memristors will allow AI to use sparse coding just like the human brain.Click To Tweet
Now, mammals use something like the memristor for image-processing, but it is a highly advanced and complex process. For highly specialized tasks that require judgment calls such as driving, cooking, or even cutting hair, we’ll need sets of specifically trained memristors. The tech may produce programs that are only capable of certain tasks, but that’s quite all right.
I don’t need my robot driver to do anything but drive, after all.
Lu’s system works very efficiently and allows a machine to recognize many different types of things by seeing just a few characteristics, and while this may be a breakthrough for the prospects of self-driving vehicles, you might want to pump the breaks a bit. The real challenge now is finding a way to scale this tech up; because a self-driving car is far better when everyone can have one, no?
The memristor might prove to be a step in the right direction, and it wouldn’t be the first one taken in the current scientific landscape where bio-mimicry is leading us to greater technological heights.
Sound Principles Found in Nature
These days we are seeing things that the world has never seen, and it’s all thanks to our scientific knowledge. With all of this super-science floating about, sometimes we can forget that nature has already refined and optimized a machine capable of highly-beneficial adaptive behaviors: the animal.
And making machines based on principles found in the animal kingdom is a tried and true way to advance our capabilities as a species.
When we wanted weapons, we made spears that were as sharp as a predator’s teeth and claws, and when we want machines that think, we made deep learning neural networks that can use a ‘library’ of data to teach itself any number of things, and we don’t yet have any idea how many things we can accomplish with those networks.
As we learn more about expanding the capabilities of our tech, we’re looking inward and toward nature for answers, and it is improving our understanding of how many things work.
A number of mental processes that go on during a trip to the store and back, for instance, doesn’t seem so complicated, but in trying to make a machine that can emulate it, we learn just how complex a simple drive down the block can be.
According to AI expert Maggie Cogen, “Artificial intelligence seeks to make computers do the sorts of things that minds can do,” so I wouldn’t be surprised if AI research comes full-circle one day and leads us to breakthroughs in psychology and education.
For now, though, we’re looking to the animal kingdom to improve AI, and I sincerely hope that works, because as I stated at the beginning of all of this: I want my self-driving car.