How AI Software Is Poised To Transform Driving As We Know It

CEO of CoCoPIE, a software solution enabling real-time AI for off-the-shelf mobile devices; Assistant Professor at Northeastern University.

AI-powered cars are, in a strange way, here and not here. The technology works—millions of miles have been traversed by true self-driving vehicles in the past decade—and it shows up in places like Tesla. Most of us don’t realize just how close AI is to becoming the future of driving. But there’s one problem: cost.

The price of AI modeling software and hardware, as well as the problem of implementation, make mainstream adoption move much slower than some of us would want. Although most of us think of AI-powered cars as using massive computers inside relatively complex vehicles, that’s not always the case. In fact, you need very little computing power to run much of the software that has the capacity to change the way we get from point A to point B.

Where’s my self-driving car?

Cars that are advertised as self-driving aren’t quite completely AI-driven. In fact, most self-driving features are tricks—dumb systems the car uses to keep us going straight and between the yellow lines.

Creating real autonomous driving requires our cars to work in L4 mode. The Society of Automotive Engineers created a six-point system (on a scale from zero to five) that describes how autonomous a vehicle really is. Most cars are L2 or “partial driving” automation. The car can control steering and acceleration/deceleration, but a human still must be in control of the car.

L4 or “high automation” cars can drive in almost any circumstances and most don’t require any human interaction. They are also geofenced, which means they can probably drive flawlessly around town but might get wonky when they head out into the country.

To achieve L4, you need to be able to create a detailed map of your environment using image segmentation and facilitate real-time response. Real-time response is absolutely essential because it allows the car to simultaneously follow a preset path—such as from your home to the store—while watching out for unexpected things like errant turns and road hazards.

By connecting an inexpensive computing module and a software AI model, you can begin to connect the real world together in a way that allows the car to drive without operator interaction. The cost of hardware is actually going down as we speak, and the software allows simple, off-the-shelf computers to gain autonomous driving powers.

How are you doing, driver?

These same AI-powered devices can make things safer for the human driver. Many dash cams are able to film the inside of the car, and by connecting a dash cam to your vehicle, your car will “know” when you’re distracted or tired. Using cheap cameras, processors and a powerful AI software model, you can turn an inexpensive gadget into a lifesaver and even embed the hardware into your car itself. This would be an essential safety measure for ride-hailing apps like Uber and Lyft that manage thousands of drivers on the roads at all hours of the day and night.

What’s next? Imagine a car that can tell you that you’re tired and that it will take over for a while, waking you when it reaches a certain point. Or imagine a sensor that can warn you—and other drivers—when you’re nervous or stressed (or, worse, drunk) and react accordingly. Again, AI plus cheap hardware can make all the difference.

You can have it in any color you want (as long as it’s black or white).

AI isn’t limited to driving. New systems that run AI in the cloud can allow you to configure your vehicle quickly and easily and even send the selected configurations to the factory where they are processed and built. The problem is that current solutions don’t operate on mobile devices and require accurate models to recognize and segment the parts.

Again, powerful software on mobile phones and other commodity hardware is pushing the envelope on how we interact with car manufacturers. These AI systems may not seem like they’re part of your smart car, but they’re definitely part of the smart car buying experience.

‘KITT, where are you?’

Once the AI knows your driving style, your habits and even what you like, why not use AI to let the car tell you about itself? By pairing an AI model with a cellphone, you can “show” your phone the car’s dashboard, and the car can respond by telling you about the various features it has to offer. This is a wild departure from the traditional “manual in a glove box,” made possible by AI on the edge of the network.

Cars, and the way we buy and use them, are getting smarter every day. Self-driving is just one step on a long road to the digital transformation of the car industry, and you can expect AI to lead the charge. What happens when everything about the automotive experience is AI-powered? That’s when we begin really living in the future.

Source