Home Connected Car Self-Driving Driverless cars explained: everything you need to know about the futuristic tech

Driverless cars explained: everything you need to know about the futuristic tech

46 min read
driverless
View original post.

Driverless cars are our transportation future, and closer than you may think.

Almost every major car manufacturer, ride-sharing service and tech company has bought into the driverless car industry. And if you take press releases at face value, we’re only a couple years away from a utopian society where your car can steer and park itself and accidents become a rarity.

But as recent fatal accidents involving self-driving cars have shown, the technology that cars use to spot pedestrians and avoid collisions still has lots of maturing to do.

With more and more companies apply for licenses to test driverless cars on public roads, we’re breaking down how companies like Google, Uber, Tesla and others train their cars’ artificial intelligences to see the road—and which AIs might have a blind spot.

We’ve gathered the latest details on which countries allow public driverless car testing, which companies are developing the smartest AI models, and what the future of the driverless car industry could bring in the next few years.

What are driverless cars?

Simply put, a truly driverless car must be capable of navigating to a destination, avoiding obstacles, and parking without any human intervention.

To accomplish this, a driverless car must have an artificial intelligence system that senses its surroundings, processes the visual data to determine how to avoid collisions, operates car machinery like the steering and brake, and uses GPS to track the car’s current location and destination. Without an AI, cars cannot be truly driverless.

Companies like Google’s Waymo put have put its AI inside virtual cars and have the vehicles ‘drive’ billions of virtual miles, throwing every perceivable obstacle at the cars to see how they respond.

The AI learns what actions lead to crashes, and slowly learns how it should drive on real roads.

To perceive the visual surroundings, most self-driving cars have some combination of three visual systems: video cameras, radar and lidar.

The AI synthesizes the data from these different systems to fully map out its surroundings and watch out for unexpected obstacles.

Most driverless cars require all three: AIs require visual cameras and deep learning software to interpret objects like street lights and stop signs, and while radar catches most obstacles instantly, it’s not as good as spotting smaller obstacles as lidar.

Still, some vehicles with autonomous capabilities like Tesla’s Model 3 don’t use lidar; Elon Musk famously called it an overly-expensive “crutch”, and that cameras and radar should suffice.

One thing to consider: the Model 3 along with pretty much every other “self-driving car” currently out there, aren’t truly “driverless”.

Most people tend to use terms like “driverless”, “autonomous” and “self-driving” as interchangeable.

But there are significant differences in the tech required for an “autonomous” AI that can only handle highways and a truly “driverless” or “self-driving” car that doesn’t even need a steering wheel or human operator to park or navigate.

Some car companies tend to fog the issue by claiming their cruise control tech for driving straight and avoiding obstacles is “self-driving”.

Mercedez-Benz actually had to pull ads that claimed its 2017 E-Class was a “vehicle that could drive itself.”

But until AI tech is sophisticated enough to drive somewhere like a school crossing without any danger to pedestrians, governments won’t allow cars to drive without a human behind the wheel.

Why should this matter to you? Because some drivers are feeling safe enough to leave the driver’s seat while their car’s in motion, putting pedestrians (and themselves) at risk. It’s vitally important that the autonomous vs driverless distinction become more clear to the public.

So, while we’re covering autonomous cars in this piece, don’t mistake them for being “driverless”; most of them have at least a few years before their AIs can properly navigate the world without a human crutch.

Why driverless cars?

For commuters, the answer is obvious: a chance to catch some extra shut-eye, get work done or watch Netflix instead of spending hours navigating through traffic. But why have companies invested an estimated $80 billion and years of work into this technology?

For starters, it could simply be a case of jumping on the bandwagon. Pretty much every major car company has developed or implemented some kind of autopilot technology into their cars. Not having that tech available could make a brand look out of date.

But at least some companies have bold business plans for self-driving tech beyond just fitting in with everyone else.

Most car brands are very concerned with their crash safety ratings. If driverless car tech will truly reduce the rate of accidents, car companies will want to push this tech forward. AI safety ratings could even become a future metric for prospective car buyers to look at.

Ride-sharing services like Uber and Lyft, meanwhile, plan to make their taxis driver free, which would eventually mean not having to pay human drivers.

In January, Uber CEO Dara Khosrowshahi said he wanted to have self-driving taxis picking up passengers by 2019, and that 20% or more of Uber’s fleet could be driverless.

Other companies like Ford hope to incorporate their cars into city-wide networks that will track traffic conditions and available parking, so the company’s self-driving cars will reach destinations faster than other cars.

Then, of course, Ford will sell their self-driving cars as a service to delivery or ride-sharing companies; Ford has already partnered with Domino’s and Postmates to deliver packages and pizza in a car that’s not actually self-driving, but pretends to be in order to gauge the public’s reaction.

Most of these companies don’t want consumers actually buying their self-driving cars.

But, at least one car industry expert claimed that car companies want their driverless tech to be a “regularly recurring subscription model”, where customers, even used-car buyers, have to keep paying for the right to not drive.

Whatever the reasons, these companies have invested too much money in driverless car AIs to stop now, despite the fact that many countries haven’t fully approved the use of self-driving cars yet yet. Businesses clearly seem to think it’s only a matter of time before driverless cars are on the road.

Where are driverless cars?

While self-driving car companies have convinced many state and national governments to let them test their AIs, nearly all governments strictly limit the cars from driving outside of testing tracks.

In the United States, 33 states have enacted legislation to allow for limited self-driving tests, but only a few states and cities let AIs be in control on public roads—and even then almost always with strict human oversight at all times.

The exception to this rule is Phoenix, Arizona, where Waymo has been testingself-driving cars without safety drivers on public roads.

Uber was also testing self-driving cars in Arizona until a high-profile fatal accident led to the state’s governor suspending Uber’s testing privileges there indefinitely.

California is another hot spot for self-driving cars, both because Silicon Valley hosts so many tech companies and because California no longer requires a human behind the wheel if companies can prove their AI is up to the task.

Cities in the US where you’re most likely to spot driverless cars include Mountain View and San Francisco, California; Phoenix, A

AtlantaPittsburgMiamiAustin, Detroit and New York City.

Europe, home to several huge car manufacturers, has many receptive countries that allow for limited driverless testing.

Germany recently approved Volkswagen to begin testing self-parking cars at the Hamburg airport.

Volvo is testing driverless cars and buses in Stockholm, Sweden. In the Netherlands, Amber Mobility plans to launch a Zipcar-like service of electric driverless cars in several Dutch cities in mid-2018.

In the United Kingdom, however, the government recently initiated the UK Autodrive initiative to push autonomous innovation, but, at the same time, the government is also conducting a three-year review of self-driving technology’s safety implications, and haven’t approved testing on public roads yet.

Australia, by contrast, has begun some public testing, but some reports say the country is lagging behind other countries in scale.

In Asia, countries like China, Japan and Singapore have enabled companies to begin testing self-driving taxis, but always with a human behind the wheel. Uber rival Didi Chuxing is one company leading China’s push for driverless tech.

As for autonomous tech found in cars like Tesla? You can find that in pretty much every nation, although most road laws dictate that drivers keep their hands on the wheel and eyes on the road at all times.

So, who’s making driverless cars?

The answer: Everyone!

Okay, that’s not entirely true, and you probably want more details than that.

Almost all of the top-selling car brands in the US—FordGMToyotaHondaVolkswagenNissanVolvoBMW and more—have been working on driverless cars for years, often in collaboration with tech developers like Nvidia and Intel.

These companies are also selling their cars to dedicated self-driving companies, like Google’s Waymo and Uber’s Advanced Technology Group (ATP), which then install their own AI tech.

We’ve got the breakdown on the biggest players in the driverless car space today, and which of them look most likely to achieve truly driverless cars in the near future.

Google’s Driverless Cars

Waymo, the self-driving division of Google’s parent company, Alphabet, was formally launched in late 2016, but its self-driving tech has been in development since 2009.

And that near-decade of work has paid off in arguably the most reliable driverless car we’ve seen to date.

Disengagement—when a human driver has to take control of a self-driving car—is the primary metric by which automakers gauge their AI’s technical skill. And Waymo’s cars lead the pack: Waymo self-driving cars disengage 0.18 times every thousand miles.

For context, if a Waymo car drove across the United States and back, a human would on average have to intervene one time. Only GM’s self-driving cars come close to that level of disengagement, averaging about 1,000 fewer miles per disengagement.

How has Waymo’s team achieve this level of reliability? With a powerful system of six lidar sensors that instantly detect any potential hazards, and a deep learning system sophisticated enough to respond instantly to obstacles and weather hazards.

Waymo collects its lidar, radar and camera feed information into an aggregate map of the surrounding road, which the company calls x-view.

The video above shows a glamorized version of how x-view can detect people and avoid accidents.

Waymo’s cars have driven 5 million miles on public roads thus far, along with 2.7 billion virtual miles inside of traffic simulators. And Waymo hopes to add to the first number in the next couple of years, as it rigs 20,000 new Jaguar I-Pace carsand thousands of Fiat Chrysler minivans with Waymo AI tech built in.


Waymo’s bold goal is to launch a “driverless ride-hailing service” in Phoenix this year, and eventually expand nationwide. We’ll have to wait and see if Uber’s recent fatal accident in Arizona halts Waymo’s plans, however.

Uber’s driverless cars

Uber’s late start (2015) to the self-driving game hasn’t stopped the ride-sharing company from zealously testing its AI tech on public roads, hoping to beat Waymo to the punch and start its own driverless taxi service.

After purchasing Otto, a self-driving truck company in 2015, Uber’s ATP developed its own system of cameras, radar and lidar to track obstacles, using a Nvidia GPU to power its AI tech.

ATP reportedly settled on just one lidar sensor, compared to Waymo’s six, to install on its 24,000 Volvo XC90 SUVs.

Uber’s self-drivings car have driven over 1 million miles on public roads thus far, though its disengagement statistics don’t stack up to Waymo’s: Uber apparently only makes it 13 miles on average before a human must intervene.

This inconsistency has led to several high-profile accidents and, most recently, after a fatal accident in Arizona earlier this year, governor Doug Ducey suspended Uber’s ability to test self-driving cars in the state, and Nvidia voluntarily suspended its own separate self-driving tests.

In light of the accident and subsequent fall-out, Uber’s contributions to the driverless car industry have been overshadowed.

Work that Uber had done included patenting a way to prevent motion sickness in passengers with a “Sensory Simulation System” that would adjust seats, air flow and in-car lighting to make riders more comfortable.

In another patent, Uber outlined how its cars could signal pedestrians or cyclists with flashing lights or a bumper text display—”intention outlets” that would help cars feel less inscrutable and difficult to predict.

What’s more, Uber has developed an autonomous truck service that will making freighting goods across the country much easier for truck drivers.

Despite the work that it’s done in the self-driving car space, Uber has a big uphill battle before the public trusts its autonomous vehicles again.

Tesla’s driverless cars

Tesla’s Model SX and 3 cars all feature the latest version of Autopilot, a sensor system of cameras, sonar and radar built for autonomous driving on highways.

Tesla’s AI can perform tasks like preemptively shift lanes before an exit or to avoid slower traffic, and can autosteer around more windy highways. Once you leave the freeway, your car will warn you to take control of steering.

As of early 2016, Tesla owners had allegedly driven hundred of millions of miles in Autopilot mode. And because Tesla scrapes data from all of its cars, it’s able to gather information on apparent errors to improve Autopilot over time. That dwarfs the mere millions of public road miles that most self-driving cars have achieved.

Of course, Tesla’s miles are autonomous, not driverless.

Tesla does sell models with “full self-driving capability” on its website, but these models apparently have only double the cameras as a regular Tesla and no other major changes.

Moreover, Tesla admits that enabling this mode would require “extensive software validation and regulatory approval” that isn’t yet available.

Still, many drivers tend to treat Autopilot like a self-driving mode, which has led to serious accidents in the past. Recently, a Tesla Model X driver was killed when his car crashed after he ignored the Autopilot’s warnings to assume control of the vehicle. The National Transportation Safety Board is still investigating the crash.

Aside from some other high-profile crashes, Tesla insists that its Autopilot and Autosteer tech generally lead to a 40-50% reduction in accidents. The below tweet shows how its tech can pick up on potential hazards most humans might miss.

For now, Tesla hasn’t announced any recent news on true driverless tech, and no one has spotted any self-driving patents by the company, either.

It’s unclear if Tesla playing things close to the chest, or if it’s content sticking with what it’s done so far while other companies duke it out over more challenging AI goals.

The other major players

Outside of these three major players, many other companies are maneuvering to start public testing, or even launch for-profit driverless car services, in the next few years.

General Motors, the runner-up to Waymo in AI reliability, plans to start testing its cars in Manhattan this year.

New York is something of an Everest for self-driving companies to climb: building an AI capable of navigating the city’s traffic and hoards of pedestrians is no easy task.

GM’s fully automated Chevy Volts each have a $5 million insurance policy for any potential crashes, and can’t enter any school or construction zones. If the cars can pass this gauntlet, GM’s AI could be powerful enough for the Chevy Cruise AV, a truly driverless car without a steering wheel or gas pedal.

Volkswagen, conversely, is braving the chaotic battleground known as parking garages for its testing.

At the Hamburg Airport in Germany, VW car owners can simply drop off their cars in front of the garage and activate a smartphone app; the car then self-drives to a free parking space, using its GPS and cameras to navigate.

Eventually, VW has designs to make your driverless car maintain itself, and even do your chores. The company stated how its cars will be able to speak with city systems to find free parking, or drive themselves to gas stations or car washes for service.

Other big name car companies haven’t made their plans public for driverless cars, but do have dates in mind for when their AI tech will be ready.

Hyundai hopes to have its cars fully driverless on the road by 2021, and Fordalso aims to have its driverless AI and traffic-tracking technology up and running in the same year.

Meanwhile, Google’s rivals in the smartphone industry also have aspirations to take the search giant on in the self-driving industry.

Apple has been keeping its self-driving car tests under wraps, but a recent patent showed Apple’s plans to install VR devices into their driverless cars to entertain passengers.

However, when people spotted Apple’s driverless Lexuses in action, the vehicles had no proprietary or custom technology visible; so it’s unclear how close Apple is to taking its driverless cars public.

Samsung also recently got permission from the California DMV to test autonomous vehicles. And even Huawei has jumped into the game, showing off a self-driving car earlier this year that ran entirely off of camera data from a smartphone.

With so many companies hoping to launch self-driving services and ramp up testing in the next couple of years, driverless car tech must be up to the challenge to avoid a rise in accidents as a result.

Both Uber and Tesla have recently been embroiled in scandals surrounding their self-driving AI after two fatal accidents this year. Any more major accidents, and trust in driverless cars could dissipate.

Below, we’ve laid out the most high-profile accidents to take place in the driverless car industry so far.

After this, you’ll find our predictions how the industry could grow in the next few years—if accidents don’t derail it entirely.

Self-driving car accidents

In 2016, when Autopilot was still newly implemented technology, a Tesla enthusiast fatally crashed into a trailer while Autopilot was engaged.

At the time, there was awareness that Autopilot had trouble picking up trailers on its cameras, but nothing had been done to fix the issue before the crash.

The incident was investigated by the US’s National Transportation Safety Board (NTSB), which initially said Tesla’s AI wasn’t at fault but eventually claimed in 2017 that Autopilot’s “operational limitations” played a role in the accident.

The agency warned that drivers using the system became too complacent to respond to any potential threats.

That pattern would somewhat repeat itself in a fatal 2018 accident, when a Tesla Model X driver crashed into a concrete barrier while using Autopilot.

According to Tesla, “The driver had received several visual and one audible hands-on warning earlier in the drive and the driver’s hands were not detected on the wheel for six seconds prior to the collision”.

The NTSB is also investigating this incident, and expressed displeasure that Tesla released its own results of the crash before the NTSB could publicly make its own statement. Tesla CEO Elon Musk claimed he had a duty to tell his customers the truth for safety reasons.

Prior to this accident, an Uber car with driverless technology struck a pedestrianas she walked outside of a crosswalk. This fatal collision led to Uber suspending all of its self-driving operations indefinitely.

As with Tesla, the NTSB investigation of the crash is still ongoing.

As for Google’s most high-profile incident, it happened in March 2016 when a self-driving Lexus SUV attempted to make a turn in front of a bus, with the car’s AI assuming the bus would slow down to allow it to do so.

However, the bus didn’t stop, and the Google self-driving car struck the bus’s side at 2 mph.

In its monthly DMV report, Google detailed the crash, and said it had adjusted its AI’s parameters to recognize that bus drivers are less likely to give right-of-way.

Most recently, a self-driving Waymo minivan was involved in an accident in May 2018, in Chandler, Arizona. But in this case, Waymo’s AI was not to be blamed for the incident.

According to the Chandler Police department, a Honda sedan ran a red light, then drove into oncoming traffic to avoid another car in an intersection, swerving directly into the Waymo minivan’s path. The human driver behind the wheel suffered minor injuries.

Waymo released footage of the incident, which makes it clear that neither the AI nor the human operator could have reasonably anticipated the crash.

Local police initially claimed that Waymo’s car had been in autonomous mode at the time of the crash, but later affirmed Waymo’s assertion that the car had been in manual mode, and they stressed from the start that neither Waymo nor the SUV driver was considered at fault for the incident

Speaking with Forbes following Uber’s fatal accident, Waymo CEO John Krafcik said that, “We’re very confident that our car could have handled that situation.”

Waymo will probably face significant backlash if it does face a serious accident of its own after Krafcik’s bold claim.

Of course, we’ll have to wait until authorities conclude their investigations into the recent self-driving car accidents before we can fully assess how safe the tech is and what steps need to be taken to avoid future accidents.

What does the future hold?

The history of the driverless car industry has been one of bold promises, high-profile fiascos, and general uncertainty about the future.

It’s truly unclear whether governments will ever let self-driving cars operate without a human operator on a national level, though it seems we are steadily moving in that direction.

A research team found that deep learning networks found in self-driving cars are prone to make thousands of incorrect choices when faced with tricky scenarios.

The researchers are hoping to develop a more complete test for self-driving car companies to check whether their AIs can navigate these problems. But, in the meantime, more high-profile accidents could be in store.

However, while accidents will play a big role in the industry’s prospects, perhaps the most important issue will be whether self-driving cars prove to be safe not just from AI malfunctions, but also malicious AI attacks.

A recent report called The Malicious Use of Artificial Intelligence, written by academic researchers and Elon Musk’s OpenAI watchdog group, detailed how hackers could infiltrate the AI of a self-driving network and cause cars to ignore safety laws.

Without protections in place, driverless cars could even become weaponized for potential attacks. The researchers recommended that companies work with one another and with lawmakers to preempt potential hacking vulnerabilities.

Will economic rivals like Waymo and Uber be willing to share such data, or will they hoard it? One can hope that companies will see the benefits of working together for the well-being of all.

If self-driving cars do take off, though, we can expect a future where companies rely more and more on autonomous tech, potentially at the expense of jobs. Amazon, for example, hopes to lower shipping costs by employing driverless delivery vehicles.

If anything is uncertain, it’s whether you or I will own self-driving cars of our own. A collection of ride-sharing companies—ZipCar, Uber, Citymapper, Lyft and BlaBlaCar—all released a policy document recommending that “autonomous vehicles (AVS) in dense urban areas should be operated only in shared fleets.”

It’s possible that self-driving car companies will continue to lobby governments for “shared fleet” exclusivity, so that you can only subscribe to their self-driving services instead of owning your own vehicle.

Of course, car manufacturers like GM and Ford will likely want to sell their self-driving cars to consumers directly, so they might lobby against such proposals.

Ultimately, with billions of dollars invested, we believe these companies will likely make driverless cars a commonplace reality within the next decade—though the road there might be littered with legislative speed bumps and public distrust.

Regardless, get ready for future generations to roll their eyes when you talk about how, back in your day, you had to drive to work yourself.

  • Everything you need to know about EVs

Leave a Reply

Check Also

What are these “levels” of autonomous vehicles?

View original post.As automated and autonomous vehicles become more common on U.S. roads, …