Improving Existing Cars

Overview - The Incremental Approach

• “I like to drive cars,” says Mark Reuss, product development chief at General Motors, “so this is a little funny.”
  
  Not funny-ha-ha, Reuss clarifies, but funny-odd. He’s sitting in the driver’s seat, with his hands on his thighs and his feet on the floor of a big Cadillac that’s driving itself around a banked oval.
  
  Reuss is at GM’s 4,000-acre proving ground in rural Michigan, hidden from the public behind locked gates, tall trees, and security befitting a prison. The company’s been debugging its cars here since 1924. It’s a brilliant, sunny autumn afternoon—a nice day for being driven. Dozens of tests are going on, though it appears that his is the only one where no one’s holding the wheel.
  
  Reuss is on edge. He forces a nervous laugh as the car takes itself up to 70 miles per hour. If he has any fast-twitch impulses rocketing across the synapses of his brain—Take the wheel, damn it!—he doesn’t give in.
  
  “This is the cat’s meow,” he says.
  
  Cadillac, which was two decades old when flappers were saying things like “the cat’s meow,” will be the first GM make to come with Super Cruise, the company’s most ambitious technological foray since automatic transmission. The system isn’t fully autonomous. Pairing adaptive cruise control with lane-centering technology, it will allow drivers, or whatever they’re called in the future, to let the car take over only on the highway. It will also, if all goes according to plan, propel GM into a multibillion-dollar race for the future of human mobility.
  
  The question is whether GM can get to the future on time. Super Cruise won’t hit the market until 2017. Elon Musk has just begun offering autopilot on his Tesla Model S. Mercedes-Benz, BMW, Audi, and Volvo have similar hands-free driving systems in the works. Then there’s Google, which wants to skip the half-measures and do a full-on moonshot: totally autonomous cars that, regulators willing, won’t even come with a steering wheel or gas pedal. Google’s latest prototypes are already driving themselves around Silicon Valley, where they’re known as Koala cars because of their bulbous shape, and they may be available for purchase right around the time GM’s hands-free Caddy hits showrooms.
  
  Google, especially, has Reuss’s attention. Last year he declared it “a very serious competitive threat.” At other times, he’s been snappish: “We’re in the car business today, and they’re not,” he said over the summer. But on the GM test oval, as he rides in the Super Cruising Caddy, he’s talking peace, perhaps even alliance. “I’m not sure it’s an us-vs.-them thing,” he says. Whether or not that partnership comes to pass (they already work together on some smaller stuff), Reuss, 52 and a GM lifer, says it’s imperative that he ...
  
  Suddenly a crescent-shaped light on the steering wheel goes from green to red and the Caddy starts drifting in its lane. Reuss grabs the wheel. A test car just whooshed by on the left, and another slowed down on the right—too much traffic for Super Cruise, making it shut down.
  
  That’s Chris Urmson, the technical director of Google’s car program, during a TED Talk in May. “The prevailing philosophy is that we’re going to take the driver-assistance systems that are in the vehicle today, and we’re going to incrementally make those better and better. And eventually we’ll get to this point where we have self-driving cars.” Urmson’s speech, viewed 1.4 million times at ted.com, is classic Silicon Valley disdain for Detroit.
  
  That kind of talk annoys Reuss, who maintains that GM can disrupt with the best of them. He uses the term 14 times the afternoon of the test drive. How to pay for that disruption is the challenge. Google, deriving profits from its ubiquitous search engine, one of the most profitable businesses ever devised, has the luxury of taking a clean-sheet approach. It has a profit margin of about 22 percent and a $495 billion market capitalization—more than triple the size of GM, Ford, and Fiat Chrysler combined.
  
  GM, whose profit margin is less than half of Google’s, can’t pour endless amounts of money into developing a driverless car. (Neither company will disclose what it’s actually spent on the technology.) But it has to finance its disruption somehow. In a sense, Reuss explains, the company has no choice but to be incremental, slowly seducing commuters and selling ever more cars—at fatter profits—as the technology improves.
  
  The CT6, Cadillac’s flagship sedan, will be the first model with Super Cruise. The technology isn’t easy to spot. Hidden behind the car’s rearview mirror is a camera that identifies lane lines and objects ahead. Two short-range radars are embedded in the front bumper, and one long-range radar peeks out from behind the grille. The camera’s primary function is to keep the car centered in its lane; the radars work mostly to detect approaching objects and keep the car a set distance from traffic. All that data feeds into two computers locked in aluminum boxes tucked beneath the spare tire in the trunk. The computers analyze the sensor inputs in real time and tell the car when to accelerate, brake, and turn.
  
  Super Cruise doesn’t have one of those spinning coffee-can things on the roof, like the ones you see on Google prototypes. Those are lidar sensors—light radar, a highly precise technology that uses lasers to read objects to the millimeter. Lidar is still expensive—about $50,000 for one unit—though Velodyne, a major supplier, has said it’s taking that price down to $8,000. GM hopes to include lidar in the next-generation Super Cruise. Google is developing its own version in-house. Its adorable Koala cars have lidar rigs affixed to the hood, looking like a button nose.
  
  But the technological differences are really just the beginning of the disruption Google has planned. In Google’s world, you won’t just quit driving cars, you’ll also quit owning them. Forget about investing in an expensive and depreciating asset that sits idle 97 percent of every day. Fleets of autonomous vehicles will circulate through your town, pick you up when you summon one via smartphone—or smartwatch or brain implant or whatever—drop you off, and move on to the next fare.
  
  In other words, Google doesn’t want to sell you a consumer product, but a mobility service, says Sebastian Thrun, a Stanford professor of artificial intelligence who ran Google’s car project for four years until 2013. “Obviously, once you get into the pure services world, that’s the end of Detroit,” he says.
  
  Reuss lets out a staccato laugh at the prospect. “I don’t know about that,” he says. “That’s a pretty dramatic comment.” And yet he accepts the premise that car sales will dramatically diminish as personal transportation becomes more service-based. In his vision of the future, GM will play in both worlds, producing autonomous cars to be used as robo-taxis and human-controlled models for the diminishing segment of society that will still wants to drive. It’s a risky strategy of reinvention, and that’s why Reuss acknowledges GM will need partners to pull it off.
  
  One thing is certain: If GM stays with its current car-selling model, it’ll go out of business. “Yep, we’re done,” he says of sticking to business as usual. “I like being in those kinds of situations. It’s kind of like Apollo 13.” He means the part about the resourceful astronauts fixing their broken capsule while hurtling through outer space, not the part about how they never made it to the moon.
  
  Driverless Cars
  
  GM has been talking about self-driving cars since its Futurama display at the 1939 New York World’s Fair. By the 1940s, its promotional films showed families playing cards around a table while their car drove itself. In the 1950s, GM developed a turbine-powered concept car known as the Pontiac Firebird, years before the muscle car of the same name, that could be switched to autopilot after getting the OK from a control tower.
  
  In 2007, GM teamed up with Carnegie Mellon University to win the Darpa Urban Challenge, run by the U.S. Defense Department's research arm. GM’s heavily modified Chevy Tahoe successfully navigated a city course on its own to win a $2 million prize. The feat “made this very singular statement that automated driving is no longer science fiction,” says Raj Rajkumar, an engineering professor who co-directs GM’s Autonomous Driving Research Lab at Carnegie Mellon.
  
  Then came the Great Recession. By 2009, as GM was descending into bankruptcy, a small band of its engineers and researchers agitated for producing driverless cars. A much larger group of more cautious executives found the idea reckless, says John Capp, GM’s director of safety, who oversaw development of Super Cruise. “We were grasping for life jackets,” he recalls of GM’s struggle to survive. Pre-bankruptcy, the traditionalists would have won. But with the government bailout came new management, which was finally convinced to invest in self-driving technology. What persuaded them were the billions that GM’s foreign competitors, including Mercedes-Benz and Toyota, were committing to similar research.
  
  And of course there was that other competitor. “There was a sense in Detroit that people at Google were going to do something foolish,” says Chris Gerdes, director of Stanford’s Center for Automotive Research, who works with automakers on driverless-car research. “But there are a lot of smart folks at Google.”
  
  If Reuss used to be dismissive about Google, he isn’t at the test track. “I love the company,” he gushes. “I love the people in it.” (This coming from a man not known for expressing much affection in the workplace: Reuss once sent an e-mail to an underling to express his displeasure after the employee used the milquetoast word “competitive” to describe GM’s models. Reuss’s subject line: “Vomit.”) GM already incorporates the Android Auto touchscreen infotainment system from Google into its dashboards. “I love working with them,” he says. “I think they feel the same way.”
  
  He wants to take this affection to the next level. “We make cars, we know how to make cars,” he says. “They’ve got great technical capabilities. We are very interested in how those two might work together.”
  
  Google declined to discuss whether it’s interested in teaming with GM on autonomous cars. However, co-founder Sergey Brin said in October, “We are really focused on working with partners” and added that they could include major automakers. Clearly, the courtship is in its early stages.
  
  Super Cruise is acting up again. After a demonstration of how it can automatically slow the Caddy from 70 mph to 40 mph and remain centered in a lane, the car drifts slightly and the steering wheel light turns red. System shutdown.
  
  The problem this time is glare from the sun, which blinded the rearview mirror camera. It’s a problem Reuss and his team have seen before.
  
  Safety comes first, of course, but if the autopilot requires the driver to take the wheel too often, then customers won’t see any benefit and won’t pay up for it. And GM is banking on Super Cruise to further boost profits, which have risen rapidly since the company emerged from bankruptcy six years ago. Reuss says GM is intent on expanding its profit margin above 10 percent, more than double what was considered healthy in the pre-bankruptcy days. Finding the right balance between human and machine control takes a lot of work. “It’s about, how much feedback is enough without being annoying?” Reuss says. To speed things up, GM simplified the system. It will operate only on highways, so it doesn’t have to deal with crossing traffic, stoplights, or children playing along the road. The company also ditched a plan for an automatic lane-changing feature, where the driver could switch lanes at the push of a button or the car itself could even sense it needs to move over.
  
  Early in GM’s work on Super Cruise, its consumer research showed that regular drivers became comfortable with autonomous technology surprisingly quickly. While taking a spin in a Chevy Malibu equipped with an early version of the technology, one test subject after another zoned out and found other things to do, such as texting or eating. “One guy reached around and pulled a newspaper out of the back seat and just started reading it,” says GM’s Capp. “He just assumed the car was going to go along forever, perfectly safe.”
  
  So GM developed countermeasures to make sure drivers keep their eyes on the road and stay ready to take the wheel. For example, Super Cruise-equipped cars will have a camera nestled among the gauges behind the steering wheel that continuously scans the driver’s eyes and face. If the driver isn’t watching the highway, the seat begins vibrating. If that doesn’t get the driver’s attention, alarms sound and the steering wheel light turns from green to blue to red. If that still doesn’t do the trick, the system will slow the car down and, if necessary, bring it to a stop.
  
  Driverless Cars
  
  There’s nothing alarming about how long it’s taking Super Cruise to come together. The technology is hard, even by Silicon Valley standards. “We’re being especially cautious at this early stage,” Musk said in October as he introduced Tesla’s autopilot. His system can handle lane changing but requires drivers to hold the wheel at all times. None of the Koalas Google has driving on public roads are for sale. There are still many thorny issues—some technical, some ethical. For example, if a collision is unavoidable, should a driverless car be programmed to always aim for the smallest object to protect its occupant? What if that small object is a baby carriage? GM is wrestling with the same issues as it tests its self-driving Cadillac SUVs on public roads in Michigan and near Carnegie Mellon in Pittsburgh.
  
  Reuss allows that Super Cruise is a ways off from truly autonomous driving, but it’s a big step in that direction. He unzips a valise and pulls out some internal documents, which he flashes but refuses to hand over. He says they contain GM’s road map to autonomy, starting with “Driver in charge” in 2010, progressing to “Driver mostly in charge” this year, to “Car mostly in charge” in 2020, and finally “Car in charge” in 2025.
  
  The road map, Reuss says, will lead GM into the bright, post-car-ownership future, where transportation is a service and driving is for hobbyists. GM could provide propulsion systems that power Google Koalas, fleets of Uber taxis, or even Apple cars, should that company decide to make them. It will also produce and sell its own autonomous cars, though Reuss isn’t sure whether they’ll still be branded Chevys and Caddys. There’s plenty of time to figure all that out, he says, dismissing the prediction that driverless cars will be the death of Detroit. “If it were going to switch overnight, maybe that would happen,” he says, slipping his secret documents back into his valise. “But it’s not gonna switch over overnight.”

Tech Creep

• The technology which will control Autonomous Cars is already being applied to today's vehicles.
  
  Automation in cars has been around for decades - adaptive cruise control has been standard since the 1990s and automatic parallel parking was introduced in the early 2000s. But recent advancements have been spectacular. The newest BMW 7-Series, for instance, lets a driver step out of the car and have it park itself in a garage at the push of a button. Some Mercedes models can drive automatically in slow-moving traffic, and both Audi and Volvo plan to introduce such technology in the next year.
  
  This week regulators and legislators move toward mandating greater use of automation in vehicles. The National Transportation Safety Board issued a new report calling for collision avoidance systems to be included in new cars; and legislation proposed in the U.S. House and Senate would require such technology to be factored into car safety ratings.
  
  Progress will reach important milestones this year and next with the introduction of technology that lets cars drive automatically on highways. The electric car maker Tesla has said it will issue an update for its Model S sedans, giving cars with the necessary sensors the ability to take control of speed and steering on highways. Cadillac plans to offer similar technology as an optional extra on some models next year.
  
  The introduction of these systems could raise concerns about driver distraction, especially if drivers suddenly need to retake the wheel. Cadillac hasn’t yet said how a driver will hand over or retake control from the system, but Dan Flores, a GM spokesman, says: “Rest assured system failure scenarios are being comprehended in the development.”
  
  Greater automation is being propelled by interest not only from automakers but also from technology companies. Besides Google and its driverless car, Uber has signaled its interest in self-driving technology by poaching academic researchers with expertise in the area from Carnegie Mellon University, and even Apple is now rumored to be developing automated driving technology.
  
  It is both taking control away from the driver, and increasing monitoring of drivers.
  
  Data on your driving habits comes from GPS and airbag sensors, and is stored on onboard computers.
  
  This can be used to vary or refuse insurance premiums, and by the courts (English driver who killed two pedestrians was jailed for 5 years after his GPS data recorded high-speed driving in the weeks before the crash).(Basically you are being nudged into 'driving like a girl').
  
  But the key takeway here is that the technology for driverless cars is being refined and proven in conventional cars.

Car-to-Car Communication

• Hariharan Krishnan hardly looks like a street racer. With thin-rimmed glasses and a neat mustache, he reminds me of a math teacher. And yet on a sunny day last September, he was speeding, seemingly recklessly, around the parking lot at General Motors’ research center in Warren, Michigan, in a Cadillac DTS.
  
  I was in the passenger seat as Krishnan wheeled around a corner and hit the gas. A moment later a light flashed on the dashboard, there was a beeping sound, and our seats started buzzing furiously. Krishnan slammed on the brakes, and we lurched to a stop just as another car whizzed past from the left, its approach having been obscured by a large hedge. “You can see I was completely blinded,” he said calmly.
  
  The technology that warned of the impending collision will start appearing in cars in just a couple of years. Called car-to-car or vehicle-to-vehicle communication, it lets cars broadcast their position, speed, steering-wheel position, brake status, and other data to other vehicles within a few hundred meters. The other cars can use such information to build a detailed picture of what’s unfolding around them, revealing trouble that even the most careful and alert driver, or the best sensor system, would miss or fail to anticipate.
  
  Already many cars have instruments that use radar or ultrasound to detect obstacles or vehicles. But the range of these sensors is limited to a few car lengths, and they cannot see past the nearest obstruction.
  
  Car-to-car communication should also have a bigger impact than the advanced vehicle automation technologies that have been more widely heralded. Though self-driving cars could eventually improve safety, they remain imperfect and unproven, with sensors and software too easily bamboozled by poor weather, unexpected obstacles or circumstances, or complex city driving. Simply networking cars together wirelessly is likely to have a far bigger and more immediate effect on road safety.
  
  Creating a car-to-car network is still a complex challenge. The computers aboard each car process the various readings being broadcast by other vehicles 10 times every second, each time calculating the chance of an impending collision. Transmitters use a dedicated portion of wireless spectrum as well as a new wireless standard, 802.11p, to authenticate each message.
  
  Krishnan took me through several other car-to-car safety scenarios in the company’s parking lot. When he started slowly pulling into a parking spot occupied by another car, a simple alert sounded. When he attempted a risky overtaking maneuver, a warning light flashed and a voice announced: “Oncoming vehicle!”
  
  More than five million crashes occur on U.S. roads alone every year, and more than 30,000 of those are fatal. The prospect of preventing many such accidents will provide significant impetus for networking technology.
  
  Just an hour’s drive west of Warren, the town of Ann Arbor, Michigan, has done much to show how valuable car-to-car communication could be. There, between 2012 and 2014, the National Highway Traffic Safety Administration and the University of Michigan equipped nearly 3,000 cars with experimental transmitters. After studying communication records for those vehicles, NHTSA researchers concluded that the technology could prevent more than half a million accidents and more than a thousand fatalities in the United States every year. The technology stands to revolutionize the way we drive, says John Maddox, a program director at the University of Michigan’s Transportation Research Institute.
  
  Shortly after the Ann Arbor trial ended, the U.S. Department of Transportation announced that it would start drafting rules that could eventually mandate the use of car-to-car communication in new cars. The technology is also being tested in Europe and Japan.
  
  There will, of course, also be a few obstacles to navigate. GM has committed to using car-to-car communication in a 2017-model Cadillac. Those first Cadillacs will have few cars to talk to, and that will limit the value of the technology. It could still be more than a decade before vehicles that talk to each other are commonplace.
  
  

• One of China’s leading carmakers is testing technology that promises to prevent accidents and ease congestion by allowing vehicles as well as traffic signals to communicate wirelessly. Although no standard for the technology has emerged in China yet, representatives at the company say it could introduce some form of car-to-car communications in 2018, ahead of many U.S. automakers.
  
  Changan, a state owned car manufacturer based in Chongquing, in central China, is testing so-called vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) technology at its U.S. R&D center in Plymouth, Michigan. The company does not sell vehicles in the U.S. and says it has no plans to enter the U.S. market. But the fact that it is testing car-to-car technology at its U.S. facility suggests that it sees a future for it in its home country.
  
  Car-to-car technology is promoted in the U.S. and Europe as a cheap and effective way to help vehicles avoid crashes and to control traffic flow more effectively. Equipped vehicles broadcast useful information, including their location, speed, and direction of travel, and computers onboard each car use that information to identify an impending collision, and issue a warning (see “10 Breakthrough Technologies: Car-to-Car Communications”). Some companies are also developing custom communications systems to allow commercial vehicles to travel in highly efficient high-sped convoys (see “Trucks Convoy under Computer Control”).
  
  Following a successful trial of the technology involving several thousand cars around Ann Arbor, Michigan, the U.S. Department of Transportation is widely expected to issue specifications for the technology later this year. The technology will debut in a high-end Cadillac in 2017, and it may eventually be mandated for new cars in the U.S. (see “The Internet of Cars is Approaching a Crossroads”). The picture is less clear in China, where the government is researching vehicle-to-vehicle technology but has not yet given any indication of when it might be implemented.
  
  I got to ride around Ann Arbor in one of Changan’s cars, a small SUV called the CS35, fitted with vehicle-to-vehicle and vehicle-to-infrastructure technology. The SUV was equipped with a wireless transmitter and receiver connected to an Android tablet attached to the dashboard. A warning flashed when another car equipped with the technology approached along a blind intersection. Another warning came as the car traveled around a sharp curve too quickly (thanks to signals received from a roadside beacon).
  
  One challenge with car-to-car technology is that it will take a while for it to become ubiquitous. Although the Chinese car market is now the largest auto market in the world, per capita car ownership still far lower in China than in the U.S., Europe, or Japan. China also lags far behind the U.S., Europe, and Japan in terms of technology development.
  
  John Helveston, a PhD student at Carnegie Mellon University who is studying the adoption of electric vehicles in China, says that foreign carmakers—which dominate the market in China—prefer to sell older technology there. And even if domestic carmakers are interested in car-to-car systems, “it wouldn’t be that interesting if only five out of every 100 cars can communicate with each other,” he says.

The 2015 Volvo

• The sensor car is here and already delivering on much self driving. After purchasing a 2015 Volvo XC60 it took me and my wife a couple of weeks to figure out what all of the sensors due. Feeling bold, I exited my neighborhood east of Seattle, pulled on to Snoqualmie Parkway. I set the speed to 70 MPH and the sensor to maintain the largest distance between me and the car ahead. I didn't touch the brake or the accelerator for the ~190 drive from my home in Snoqualmie WA to Beaverton, OR to visit a customer. From Snoqualmie, WA: Snoqualmie Parkway (South)--> WA State Route 18 (West)-->WA SR 167 (South)-->WA SR 512 (West) --> Interstate 5 (South)--> I-405 (South) --> Oregon Hwy 26 (West) --> Exit Hwy 26 onto NW Murray Blvd where I applied the breaks for a stop light. (David James McCabe on LinkdIn)

Tesla Self-Drive Cars Are Here

• With one over-the-air update Wednesday night, Tesla Motors has brought a new breed of self-driving car to American roads.
  
  Tens of thousands of Tesla's all-electric sedan, the Model S, bought in the U.S. over the last year have already started downloading or installing "Autopilot" mode, one of the first great breakthroughs for making the kind of driverless magic seen mostly in Google-car demos.
  
  With "Autopilot," the Tesla S can steer, change lanes and drive at highway speeds with little to no help from the human behind the wheel. It can parallel park, using its banks of cameras and sensors, and slow to a stop if the driver happens to drift asleep. In the next update, it may even be able to rouse itself from its parking space and pick the driver up.
  
  It's called "Autopilot" for a reason: Like airplane pilots with takeoffs and landings, the driver will still be expected to handle much of the subtle and strange ballet that is modern driving. The human will still have to keep her hands on the wheel every few seconds, as a safety measure, and to meet state laws that demand a hand on the wheel. No naps in the driver seat.
  
  But the features it brings — as a software update, no less! — mark an incredible turning point for driverless technology, which could dramatically reduce the 33,000 deaths on U.S. highways every year, advocates of the technology say.
  
  "It will get more and more refined over time," Tesla chief executive Elon Musk said Wednesday at a press event. "Eventually, we want it to automatically have your car put itself to bed in your garage."
  
  When the Model S now is, say, cruising on the highway, its cameras, sensors and radar are already gulping in data on lane markers, the speed and location of other cars, and other important information. Before the update, the car could automatically slow down when, say, it sensed the car ahead hit the brakes, and it would also vibrate the steering wheel when the car seemed to drift out of its lane.
  
  Turning on "Autopilot" instructs the car to take an even greater level of control, steering into curves and maintaining its speed. To change lanes, the driver can hit her turn-signal bar, called a stalk, and the car will speed up a little and drift over. The driver still has to check her blind spots before hitting the signal; if the car senses interlopers, it will keep going straight.
  
  Parking also gets an upgrade. The car will now be able to, as Tesla says, "scan for a parking space, alert you when one is available, and parallel park on command." The car's back-up camera and sensors say exactly how many inches the car is away from the curb.
  
  The Model S will give back control to the driver if she grabs or turns the wheel. Alternatively, it will sound a series of increasingly annoying beeps if the driver doesn't seem to be paying attention to the wheel.
  
  If the driver doesn't snap to attention? The car flashes a warning for the driver to grab hold of the steering wheel and eventually slow to a stop where it's driving, firing off the hazard lights. (Other carmakers have similar backup plans but differing ideas on how best to jolt drivers awake.)
  
  Consider it something like a diet driverless car. It's meant for the relatively predictable expanse of highway driving, and won't work below 18 mph. It won't always be able to navigate darting pedestrians and the many other confusing obstructions on the road, which for now will remain the human's domain.
  
  Of course, that also means, as Tesla said, any accident when the car is in or out of "Autopilot" remains the liability of the driver. That may change as the cars become smarter and more assertive, but for now, driver beware.
  
  Unlike Google, which is aiming for a fully-autonomous vehicle, Tesla's approach is to gradually introduce features which take away the need for drivers to carry out certain functions.
  
  Currently there are limitations to the software which would improve over time, Mr Musk said. "If there's heavy snow it's going to be harder for the system to work, so we'd advise caution. "Essentially it's like a person - how well can a person figure out what route they should take. Over time it will be better than a person. "Long term it will be way better than a person. It never gets tired, it's never had anything to drink, it's never arguing with someone in the car. It's not distracted."

Tesla Self-Drive Cars Learning

• During the press conference for the release of the Autopilot, Tesla CEO Elon Musk referred to each Model S owners as an “expert trainer” – meaning that each driver will train the autonomous features of the system to feed the collective network intelligence of the fleet by simply driving the electric vehicle on Autopilot.
  
  He said that the system should improve everyday, but that improvements might only become noticeable every week or so by adding up. Just a few weeks after the release, Model S owners are already taking to the Tesla Motors Club forum to describe how the Autopilot is improving.
  
  A common problem with the early version of the system was that it had a tendency to try to take exits on the highway when it wasn’t supposed to, but after a few tugs on the Autopilot’s leash, trainers have corrected the issue.
  
  Model S 85D owner mobe:
  
  The first day when I was in the right lane, as I approached exit ramps, it would dive for the exit ramp. I quickly learned to apply torque to the wheel to hold the car on the interstate until I had passed the exit. Each day the system seems to have less tendency to follow the exit ramps as I pass. The last two days it only gave a momentary wiggle and moved over maybe six inches towards the exit ramp then it recovered and moved on down the road. This morning it gave only a very slight hesitation, so little that I did not have to correct it at all. I find it remarkable that it is improving this rapidly.
  
  Tesla doesn’t recommend using the Autopilot on roads with aggressive curves, but of course it doesn’t stop owners from trying and apparently the system is learning to handle the sharp curves pretty fast:
  
  Model S P85D owner commasign:
  
  I noticed that on sharply curved ramp connecting I-80 west with CA-113 north in Davis, the first time it took the curve at full speed and wasn’t able to stay in lane resulting in a “take control immediately” alert. After a few more times on this curve with firm pressure on the steering wheel it’s now learned to slow down and today had no issue taking the curve. Definitely learning.
  
  A Model S owner by the username Khatsalano kept a count of how many times he had to “rescue” (meaning taking control after an alert) his Model S while using the Autopilot on his daily commute. He counted 6 “rescues” on his first day, by the fourth day of using the system on his 23.5 miles commute, he only had to take control over once.
  
  Sillydriver explains how the Autopilot is learning his commute:
  
  AP is definitely is learning. Each weekday morning I drive home on Route 50 in the Virginia countryside after dropping my kid at school. Watching the display while driving manually last week, I could see that AP lost lock on the road where Route 50 drops from a 4 lane divided highway to 2 lanes. Where the left lane merges into the right, the dashed line disappears for a while. Later on in the 2 lane section, a right turn lane comes off during a left hand curve in the road. I could see the projected course bend right and then snap back left: I had been using AP on the 4 lane stretch before the lane drop and was glad I had turned it off.
  
  During the last couple days I have seen the projected course tracking correctly in both problem locations. The time of day, lighting and weather has been the same every time, and there were never any cars ahead to follow (I’m traveling against morning traffic). Then this morning I tried AP the whole length of my trip on 50 and it tracked perfectly.
  
  Last week, Tesla released the Autopilot in Europe and Asia. While announcing the regulatory approval in the new markets, Musk also announced that on top of the fleet learning, the company will “soon” release a 1.01 update to the Autopilot software to further improve the system’s capacity to self-improve and handle curves, as well as lane changes.