I was sitting in the backseat of a Volkswagen Touareg.
A computer tucked away in the hatch hummed along, and an LCD panel gleamed as bright as the California sun. My friend sat in the passenger seat and his daughter was right next to me. An engineer was in the driver’s seat, but no one was driving. The car negotiated around a closed-off parking lot by itself.
That was ten years ago when the actual test took place. Since then, there have been a few ups and downs with autonomous cars. This summer, a driver fatally crashed into a semi-truck while driving autonomously in a Tesla Model S sedan. Google has tested their robotic cars in San Francisco for several years, and they’ve had a few close calls and at least one crash at low speed.
Now, Ford has announced they will start selling production-ready self-driving cars by 2021. These won’t be laboratory experiments on wheels; the plan is for customers to use these cars in their daily commutes. They’ve invested heavily in a new research initiative. Meanwhile, Uber has announced they will start using a fleet of Volvo XC90 vehicles in Pittsburgh soon, although these autonomous cars will be “human assisted” (a driver will keep their fingers on the wheel). Audi has already shown they can drive a car autonomously from the Bay Area all the way down to Las Vegas. Cadillac and several other automakers are busy testing their own robotic tech.
What’s taking so long? In the past year, I’ve had two important realizations about the slow pace of innovation when it comes to fully autonomous cars, which is the true end goal and not just some robotic assistance. These vehicles don’t even need a steering wheel or brakes. They can scan the road constantly, adjusting speed and steering, and they will be smarter than most human drivers (e.g, that guy texting on the highway).
My first realization is that full self-driving capability is really outside of the control of the car itself. Imagine turning over the controls to a robot. In that parking lot ten years ago, the car could drive around easily on its own. The LIDAR sensors they used could scan for any obstruction. Yet, the engineer was quick to point out even then that there are too many hairy situations for robotic cars. In San Francisco, Google can easily control the drive path and stick to the main arteries. But what about the Golden Gate Bridge and the fact that people tend to drive unusually slow, sometimes looking for whales or gazing over the horizon?
A sensor doesn’t know that. It’s when the bridge itself connects to the car, alerting the computer about sudden slowdowns, that this will become smooth enough for a company like Ford that have a reputation at stake.
Another example: In some areas of San Francisco, there are so many bikes that you have to stay hyper-vigilant when you drive. I was one of those bicyclists a few weeks ago. Today, you can train a car to look for bikes, to know which routes are congested with bikes, and to stop for a bike. Great.
What you can’t do is connect the car to a bike. I’m hoping, in five years, the internet of things hits such a major growth period that many of these bikes have a simple clip-on transmitter that alerts a self-driving car that you are about to take a quick right turn on Embarcadero. It will save your life.
My other realization is that this is all about risk assessment. Tesla is willing to accept the consequences for the sake of innovation. It looks like Uber is also willing to do a limited trial. Both companies are trendy and cool.
But what about Volvo? The Swedish company has been around for 89 years. They have 88,464 employees. There’s a reason major automakers innovate quickly (to stay ahead of the competition) and move to production slowly (to avoid a catastrophic publicity nightmare).
Ford has almost 200,000 employees. I’ve mentioned this many times to friends and family, but it gives you a clear picture of the industry when you realize that Ford sells about 3,200 trucks per day and Tesla sells about 3,700 vehicles from their entire fleet per month. Ford, Volvo, Audi and the major automakers need to make sure they get this right. The cars need to be aware of the road conditions, the position of other cars, the fact that people do whale-watching on the Golden Gate bridge, the sudden shifts in traffic conditions that occur at 5 p.m. in Des Moines, and every other possible scenario. That takes time, industry cooperation, IoT advancements, and many other steps.
To be honest, a computer needs to be even smarter than a human driver. I found that out this morning when someone didn’t yield at a round-about. Oops. Real robotic intelligence requires skill, patience, cooperation, and a willingness to accept risk. It also takes time.
This article is published as part of the IDG Contributor Network. Want to Join?