Internet Takes The Wheel Inside Driverless Cars
Fully automated vehicles are teetering on the edge of commercial viability.
As researchers around the globe continue to tinker with autonomous driving software, they're also anticipating its potential impact. Handing the keys over to algorithms means our cars will, in effect, become an information technology. But unlike laptops and smartphones, connected cars will alter the world around us.
Aside from the very-much-in-beta fully automated vehicles (AVs) currently being road-tested by the likes of Google, technologies that allow cars to operate at least somewhat independently have been with us for years and, in some cases, decades.
In 2013, the US National Highway Traffic Safety Administration (NHTSA) released a blueprint outlining how advanced forms of automation should be introduced to public roads. It included five levels of autonomy ranging from Level 0 ("No Automation") to Level 4 ("Full Self-Driving Automation").
Most cars on the road today are either Level 1 (the driver controls everything) or Level 2 (which incorporates newer bells and whistles like adaptive cruise control and automatic lane centering). Level 3 provides limited self-driving automation; drivers are still expected to commandeer navigation at some points along the journey. Once we reach Level 4, passengers just get in and say "Hey Siri, take me to grandma's house." That's what Google is working on; its prototype has no steering wheel or pedals though the Little Tikesesque AVs currently cruising California streets require them by law for now.
Before anything approaching Level 4 is let loose in the wild, though, a not-insubstantial number of legal, ethical, and technical issues need to be addressed, the sensors on Google's cars, for example, reportedly still have trouble discerning between a bag blowing in the wind and a deer galloping into oncoming traffic. Still, it's a good bet that most people reading this will see an autonomous car on a street near them in their lifetime.
Car makers including Tesla, Toyota, and Volvo have already promised to deliver fully autonomous vehicles by the end of this decade. The question is no longer, "Is it possible?" but rather, "How long until it's available"?
"I see two kinds of scenarios," says Professor Raj Rajkumar, the co-director of the General Motors-Carnegie Mellon Collaborative Research Lab. "First is that I can see vehicles being deployed in restrictive scenarios where the road is clear of pedestrians and bicyclists, and the vehicle could stop only at designated places. People would get on or off from the vehicle at specific locations—just like a shuttle, for example. I can see that happening in two to three years' time."
Rajkumar also suggests that we might see a continued piecemeal introduction of features (for example, GM's super cruise or Mercedes-Benz's steering assist). This steady accretion, he says, can bring us to Level 4 automation in "about 10 years."
"I think we are on the right trajectory. The technology has been demonstrated in relatively constrained situations, for example, in regions where there is no heavy rain or snow," Rajkumar points out (though earlier this year, Ford started testing its autonomous vehicles in snow). "Then we have also had a bunch of recent incidents where the interactions between human drivers and self-driving vehicles are still being fleshed out."
By "fleshed out," Rajkumar was diplomatically alluding to a number of recent accidents involving AVs, which were mostly caused by the human-navigated vehicles sharing the road with them. Our interview took place a few weeks before the public learned about the NHTSA's investigation into the first known fatality involving self-driving technology.
In that case, an early adopter placed a little too much faith in the semi-autonomous Autopilot feature on his Tesla Model S. He trusted that his car would be able to decipher between the brightly lit sky on the horizon and the white side of a tractor trailer running perpendicular across the highway. It was not.
Not to relegate a person's death to a statistic, but history will probably view this accident as one awful step back before we take several gigantic leaps forward in regards to public safety. In contrast to much of the media frenzy surrounding the deadly accident, this incident reinforces the need for more automation on the roads, not less.
The Most Dangerous Thing You Do Every Day
Safety advocates, regulators, and the car industry are quick to point out that America's roads are safer than at any point in history. Road deaths have been nearly halved over the past four decades, dropping from 53,000 in 1970 to 33,000 in 2014, which is even more impressive when you consider the population has swelled by half, and we've tripled the number of miles driven annually.
While we've come a long way toward mitigating the carnage, we shouldn't lose sight of the fact that a stadium's worth of people die each year on US roads (plus more than a million more around the world). Even if you're the most attentive hands-at-ten-and-two driver, who always obeys the speed limit and never gets behind the wheel after imbibing even a sip, there's a good chance some other driver sharing your stretch of road isn't as responsible.
We could continue implementing partial solutions, such as seatbelt laws, crumple zones, and median barricades, or we could accept the underlying problem: humanity. The good news (from a public health perspective) is that humans are a problem that is easily correctable, technologically speaking.
Like any complex machine, AVs are not the result of a single technical breakthrough. Most current models incorporate several kinds of sensors (optical cameras, RADAR, LiDAR), which provide a steady stream of real-time data to increasingly "wise" algorithms. Specifically, AVs employ "machine learning" algorithms. Machine learning is a subset of artificial intelligence that allows computers to react to novel scenarios they weren't specifically programmed to encounter (as no program could possibly anticipate every road eventuality).
The path to self-driving critical mass will probably be messy, but hopefully not deadly, especially as we enter the transitional time when the roads are shared by cars navigated by neurons and those by algorithms. To that end, there is a public-private effort to develop solutions such as vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication technologies that would theoretically prevent accidents like the one involving Tesla's Autopilot. But more important, they'll work in conjunction with an AV's onboard tech to ensure it operates predictably, efficiently, and safely.
Beyond handing potentially consequential decisions to robots (and yes, AVs are robots), the very prospect of them infringing on our turf elicits a knee-jerk unease among many people. But history has shown that society inevitably learns to embrace the benefits of the non-sentient entities in its midst, be they faceless ATMs offering 24-7 convenience or the steely perfection of automated airport monorails that operate with little-to-no human oversight.
The prospect of fewer deaths may be this technology's most compelling raison d'etre, but it's far from the only benefit. AVs will open the world to people who lack access to the analog road due to financial, medical, or legal prohibitions. Even for those with the means and ability, driverless technologies will completely revolutionize the way we get from A to B.
Uber execs have been notably aggressive (as is its reputation) in their push to bring this technology to market. Uber will begin real-world tests of self-driving robo-taxis in Pittsburgh later this year via autonomous Volvo XC90s with humans in the driver's seat.
Even that most iconic of American automakers, the Ford Motor Company, has transitioned from viewing vehicles solely as things that customers buy to a more fundamental role as things that get customers from point A to B. At this year's Us Consuner Electronics Show , Ford CEO Mark Fields announced Ford's pivot towards technology-driven "transportation services," including alternative models such as ridesharing, pay-by-mile rentals, and continued investments in self-driving tech. More recently, Fields tipped self-driving robo-taxis (without pedals and steering wheels) by 2021.
In the future, entering a car might be similar to going online today: You'll be inundated with all manner of services, which run alongside easily ignorable ads and nano-targeted corporate messaging. Perhaps there will be a way to pay for a premium, ad-free experience or deploy the vehicle version of ad blockers.
Drive.ai uses deep learning to teach self-driving cars
Startup Drive.ai is revealing its product and strategy for the first time, and the autonomous driving tech company is looking not only to create the best hardware and software to enable self-driving cars, but also to make sure those cars communicate with people outside of the car in the most effective way possible.
Their approach is using deep learning across the board in its autonomous driving system, which means teaching their self-driving cars somewhat like how you’d teach a human. That involves providing a host of examples of situations, objects and scenarios and then letting the system extrapolate how the rules it learns there might apply to novel or unexpected experiences. It still means logging a huge number of driving hours to provide the system with basic information, but Carol Reiley, co-founder and president of Drive.ai, believes that it should also help their self-driving vehicles deal with edge cases better.
