英国知名汽车杂志Automobile记者Todd Lassa近日走访了神秘的Google X实验室，和无人驾驶汽车项目组的科研人员进行了深入交流，并亲身体验了紧张刺激的「无人车搭乘之旅」。下面我们跟随Todd的视角，去看看谷歌在无人驾驶汽车安全、事故处理、路测资质评估以及整车制造方面，都有哪些故事分享。
Rather than neat, Orwellian rows of identical autonomous cars in a constant, regulated flow, Google envisions a motorized mishmash not terribly different than today’s traffic, though it would be safer, cleaner, and more efficient. The company is concentrating the bulk of its efforts on solutions for urban and suburban environments, with robotaxis and Uber cars (Google is Uber’s biggest investor) joining the conventional traffic mix by the middle of the next decade. Mainstream automakers have thus far focused their efforts on step-by-step buildups of the adaptive cruise control and lane departure control, building blocks necessary to bring automation to the nation’s highway system. Everyone is baking in redundant systems—elaborate sets of sensors, cameras, and mapping technology—designed to check and recheck one other with the promise of dramatically minimizing unavoidable impacts.
Crashes will happen, but Google management does not seem terribly concerned about the question of who is ultimately responsible when the first major crashes involving driverless cars take place. (Google’s autonomous vehicles have been involved in 12 minor fender-benders since testing started in 2009 in California, although Google claims its cars and drivers were not at fault in any case.)
“Systems fail. We’re doing everything we can to make this safe. It’s going to reduce crashes, reduce injuries, but it’s going to be mixing with fallible human traffic, fallible human drivers. It’s going to have defects,” explains Google’s driverless car program chief Chris Urmson. “Hopefully, they will be far fewer than the rate of accidents, the rate of injuries that we’re seeing on the road today. The good news is that the U.S. legal system is really robust, and this question of who’s really liable is going to get sorted out.”
Google’s conventional Lexus RX crossovers and Toyota Prius sedans, the ones with the ungainly aftermarket rack-and-cone setup on their roofs, continue to roam Northern California—Google says its vehicles have traveled the equivalent of some 1.7 million miles, 1 million of them in driverless mode—while the company awaits a ruling from the California Department of Motor Vehicles that would allow it to test its new prototypes without driver controls on public roads. Google expected the DMV ruling by the beginning of this year, but governments do not move with the same alacrity as Silicon Valley. Whenever the ruling comes down, Urmson says, Google will be ready.
Florida, Nevada, Michigan, and Washington, D.C., have joined California in implementing laws allowing testing of automated cars on their roads in some form, but getting the 46 remaining states in line might become a near-term stumbling block, especially when it comes to cars without redundant controls. Richard Wallace, director of the Center for Automotive Research’s (CAR) Transportation Systems Analysis group, fears states will not all join in with similar laws. “[The National Highway Traffic Safety Administration] can bring a strong carrot or stick, tying in federal funding, but ultimately it falls to the states to decide,” he says.
在充满稀奇古怪玩意儿的Google X实验室里，藏着属于无人驾驶汽车项目的一片天地。相比其他较封闭狭小的区域，无人驾驶项目组拥有类似厂房开阔的设计。一眼望去，只见一排排头顶Lidar的试验车整装待发，侧面留有供工作人员调试系统的工作台。目前留在这里的大部分都是谷歌去年推出的无人驾驶原型车，偶尔能看见几辆雷克萨斯RX和丰田普锐斯，它们主要进行系统调试及数据采集工作。据一位首席研发工程师Jaime Waydo介绍，谷歌去年发布的这款无人驾驶电动概念车，最大载重不超过1360千克，外部车身采用碳纤维材料制成，一台电动机位于后轮靠下的位置，是一款再简单不过的后驱式电动车型，由密歇根州的一家工业机械公司Roush Industries负责制造装配。此外，这台无人驾驶电动车的最高时速为40公里/小时，锂电池综合续航约为120公里。当然，整个车子最关键的部分自然是那套传感器系统。它由位于车顶类似警笛设备内的激光雷达，车前小狗鼻子似的摄像头和前饰板下的一枚激光探头组成。据介绍，头顶处的那枚激光雷达，最远能够探测到两个标准足球场距离外一辆车的动静。
Whenever California finally clears the driverless prototypes for use, they will be classified as neighborhood electric vehicles with a maximum loaded weight of less than 3,000 pounds. (Think glorified golf cart.) Google’s prototypes—not its Lexus and Toyota test cars but vehicles built by Michigan’s Roush Industries—employ a carbon-fiber body and an electric motor of unspecified output, mounted low in the back and driving the rear wheels, says Jaime Waydo, vehicle development lead engineer. Top speed is 25 mph, and the battery pack is rated for a range of 75 miles. Sensors include a laser in the siren-look device on the roof, a camera in the puppy-dog nose, and a laser under the front fascia. The top laser “can track a car up to two football fields away,” Waydo says.
Its rounded body lines make it easier for the sensors to read a 360-degree image of the car. Inside, a narrow, horizontal screen at the cowl displays the camera’s view to passengers, and a large, open compartment for groceries and backpacks is between the cowl and the passenger footwell. So when will it be ready for general consumption? “I always answer, ‘We’re never going to put it out there until it’s safe enough.’ But the cute answer I give is I’ve got an 11-year-old son. He’s going to be 16 in five years,” and Urmson wants him to continue being a passenger instead of getting a driver’s license.
不过遗憾的是，由于谷歌不允许试乘这辆最新无人驾驶汽车，无奈之下，Todd只好搭上了一辆雷克萨斯RX 400h。项目经理Brian Torcellini按下方向盘上的一个按钮，随后车子便开始自动驾驶，依靠高精度地图即可走遍山景城。它能够自动遵守当地的交通规则，偶尔导航系统会发出类似“前方有十字路口”这样的女声提醒。不过在高速公路上，按照谷歌工程师的设定，为了保证跟车顺畅，允许无人驾驶汽车超速行驶。Todd表示，除在有自行车或其他车辆突然并入前方车道的情况发生时，这辆雷克萨斯RX 400h会猛踩刹车制动外，整个行驶过程基本还是十分平顺的。
Google won’t let us roll in a prototype, but we do get a ride in the back seat of a Lexus RX 400h. Program manager Brian Torcellini hits a button on the steering wheel, and the car does all the work, following a Google map around greater Mountain View. It obeys speed limits and traffic laws, and occasionally a female navigation voice warns of a “crosswalk ahead” and the like. On freeways, Google lets the car exceed speed limits enough to keep up with traffic. The Lexus is smooth and efficient—until a bicyclist or another car veers into its path. That’s when it slams the brakes, hard.
“The car is making checks all the time” and maintaining a safety cushion, software engineer Nathaniel Fairfield explains. “We ask, ‘Why are you slowing down?’ And there’s a reason for it. “We were testing [an autonomous Lexus] in Mountain View, and there was this woman in a powered wheelchair, with a broom, chasing a duck in a figure eight in the road,” he says. “We can take that data, and add that to the repertoire of situations that the vehicle has to test against and make sure we’re safe with women in power wheelchairs chasing ducks in roads.” So grandmas chasing ducks around in figure eights are covered, but what about the costs to consumers for all this cutting-edge technology? New technologies tend to come down in price when they’re mass-produced and there is competition. Proponents point to the same red herrings that were also trotted out before antilock brakes, electronic stability control, and external cameras became ubiquitous.
Take lidar, for example. The laser-based radar systems are a critical part of most of today’s autonomous suite of technologies. “The lidar used on the 2007 DARPA Challenge-winning GM vehicle, made by Velodyne, was a $75,000 piece of equipment,” CAR’s Wallace explains. “They’ve got one akin to that now that’s $40,000 and a smaller one that’s $10,000. Meanwhile, some of their competitors are making automotive-grade lidar, significantly smaller devices—so you may need two of them—that are in the $500 [each] range. … I think it’s going to be a modest markup in cost for tremendous benefit.”
So, by 2030 or earlier, if you believe Google and other automotive futurists, the transportation world will have changed forever, and it probably can’t come fast enough for many of today’s drivers. You only need to look around you on the road to estimate what portion would rather be texting, calling, eating breakfast or, ahem, Googling on their mobile devices.
Future Past：Autonomous Advances Through the years
1939: A scale model diorama of Norman Bel Geddes’ vision for an automated highway system highlights the Futurama ride in the General Motors pavilion at the New York World’s Fair.
1958: Chrysler offers Auto-Pilot, the first production cruise-control system, based on mechanically manipulating the throttle linkage.
1958: GM and RCA experiment with automated cars, using wire buried in pavement and magnetic pickup coils in cars. A Chevrolet Impala serves as the test car. GM later uses this technology inits Firebird III and experiments with radar in the Cadillac Cyclone dream car.
1973: Autonomous cars in Woody Allen’s sci-fi comedy “Sleeper,” set in 2173, presage the 2015 Mercedes-Benz F 015 Luxury in Motion concept.
1980s: Professor Ernst Dickmanns, an expert on artificial intelligence from the University Bundeswehr Munich, tests a “robot car” and joins the $1 billion, pan-European 1987-’95 Program for European Traffic with Highest Efficiency and Unprecedented Safety (PROMETHEUS).
1982: An autonomous Pontiac Trans Am known as KITT—short for Knight Industries Two Thousand—is the major (only?) reason to watch the “Knight Rider” television series.
1994: Dickmanns demonstrates a robotic Mercedes-Benz 500 SEL with humans in the front passenger seats for more than 620 miles on Paris multi-lane highways at speeds of up to 80 mph. The next year, his robotic Benz makes a Munich-to-Denmark round-trip, driving up to 90 minutes at a time without human intervention, at speeds up to 112 mph.
1995: A Carnegie Mellon University robotics team drives a self-steering 1990 Pontiac Trans Sport minivan dubbed Navlab 5 from Pittsburgh to Los Angeles, where Jay Leno declares himself unimpressed.
1996: A Lancia Thema, developed at Italy’s University of Parma, follows highway markings while covering nearly 1,250 miles in Italy, with 94 percent of the trip driven in autonomous mode.
1997: The federally sponsored Automated Highway System is demonstrated on I-15 in San Diego with a fleet of eight Buick LeSabres platooning via vehicle-to-vehicle communications. Additional tests feature automated Pontiac Bonnevilles, an Oldsmobile Silhouette, and full-size New Flyer city buses.
2001: The newly passed Defense Authorization Act declares that one-third of all American military ground vehicles shall be unmanned by 2015.
2004: The Defense Advanced Research Projects Agency (DARPA) offers a $1 million prize for a driverless vehicle able to complete a 142-mile course through the California and Nevada desert. Although the much-ballyhooed event draws high-powered teams from industry and academia, all 15 qualified vehicles either break or get stuck.
2005: Stanley, a driverless Volkswagen Touareg modified by a team from Stanford University, claims the second DARPA Grand Challenge—and a reward of $2 million. This time around, five of the 23 entrants complete the off-road course.
2007: Carnegie Mellon’s DARPA team, which finished second in 2005, turns the table on Stanford and claims the $2 million prize for the Urban Challenge—the final DARPA event for autonomous vehicles—with a sensor-laden Chevrolet Tahoe.
2010: An Audi TTS named Shelley (for rally legend Michèle Mouton) by its developers at Stanford makes it to the top of Pikes Peak. Over the next few years, it develops the road-racing chops to lap as quickly as race-car drivers.
2011: Unmanned electric-powered Ultra Global Personal Rapid Transit (PRT) pods begin operating on a 2.4-mile defined route from a remote parking lot to the U.K.’s Heathrow Airport.
2012: Nevada issues the first license for an autonomous car to a Toyota Prius modified by Google, whose driverless car program is headed by Sebastian Thrun and Chris Urmson, alumni, respectively, of the Stanford and Carnegie Mellon Grand Challenge teams.
Here I am: Vehicle-to-vehicle Chatter Increasing
Automakers are slowly beginning to roll out vehicle-to-vehicle (V2V) communications systems, which could potentially become a big part of the suite of safety technologies for semi- and fully automated driving. Cadillac, for example, has announced it will offer a V2V package as an option for its 2017 CTS.
V2V technology is designed to allow any connected vehicles within a certain area to talk with one another anonymously through the use of a basic application known as a “Here I Am” message that can be communicated via devices such as GPS. By monitoring each vehicle’s position, speed, and location relative to the others, connected vehicles would then be able to calculate risk from a 360-degree perspective, allowing for pre-emptive action in the form of a warning alert when the system determines a crash is imminent, for example.
“The car in the front has a camera. You can see in front of it, and then that image can be transmitted to a car that’s a hundred cars behind,” says Frank Paluch, Honda R&D Americas president. “If we can get every car talking to each other and use that information, that sure would transform our mobility, our society.”
One of the more interesting aspects of V2V tech is the potential for aftermarket solutions that could allow for quicker adoption among older cars, which is a core issue. That would come with a dollar sign attached, which the National Transportation Highway Safety Administration estimates at roughly $350 a vehicle by 2020. Despite the obstacles, the feds are high on the concept and have been studying it for more than a decade.
Google? Not so much. It says it isn’t in any hurry to incorporate V2V. “It’s going to be 50 years until the last pickup truck [without V2V] is off the road,” says Google software engineer Nathaniel Fairfield.
Who will be Google’s automotive partner?
Google says it has no plans to jump into the automaking business. Instead it wants to team up with a manufacturer that will help it mass-produce autonomous vehicles based on its prototypes built by Michigan-based Roush Industries. Initially, the cars are envisioned as urban/suburban taxis, Uber cars, and “last-mile” cars that connect bus and metro passengers with their homes and offices. We expect production models to be electric-powered, like the prototypes.
The Internet giant says only those at the very highest levels of the executive flow chart are in discussion with a potential partner or partners, and they’re not talking. That does not stop us from speculating on which companies are the best prospects for such a plum deal. They’re listed from most to least likely:
The conventional cars Google has so far converted for autonomy testing have been the Prius and a smattering of Lexus models. Since Toyota likely has already worked with Google and it has chops as the world’s biggest automaker, it’s the leader in the clubhouse. Scion aG anyone?
Mercedes-Benz’s parent is another leader in automated vehicle development. Its new Smart Forfour seems the right size, and it is rear-wheel drive like the current Google prototype.
CEO Carlos Ghosn says traffic jam pilot, which “allows the car to drive autonomously and safely in heavy, stop-and-go traffic,” will be available on several Nissan and Renault models by 2020, and the electric Leaf could make an ideal platform and package.
Before he joined Google, Chris Urmson led Carnegie Mellon University’s DARPA effort in a collaboration with GM, and together they won the 2007 DARPA Grand Challenge with a Chevrolet Tahoe. GM could amortize development of the 2017 Chevy Bolt with a Google car using the Bolt’s platform and electric powertrain.
Fiat Chrysler Automobiles
FCA CEO Sergio Marchionne is looking to hook up with a partner, and if no mainstream automakers bite, he is open to a deal with a nontraditional partner such as Google or Apple. Fiat does have a 500 EV that kind of, sort of looks like the Google prototype.
A new biography of CEO Elon Musk says he nearly sold Tesla to Google in early 2013 as the auto company faced bankruptcy. More recently, Musk said Tesla has autonomy “solved,” but it is at the bottom of our list because what Google needs most is a partner with economies of scale and deep manufacturing know-how.