Brake lights flashed. Cars inched forward, sped up, and slowed down. Drivers swooped from lane to lane, searching for the slightest of advantages in the late-afternoon rush. It was all part of a routine commute along Interstate 15—for every car except one.
Operating amid the traffic fray was an Audi SQ5 equipped with an automated driving system jointly developed by automotive suppliers Delphi and Mobileye.
During an autonomous journey that covered a little more than six miles, the car negotiated a highway merge from the on-ramp and later made a graceful exit, two challenges that many engineers consider some of the most complex in self-driving operations. Accomplishing the maneuvers at the peak of rush hour only enhanced the feat.
Among the dozen or so self-driving demonstrations taking place in Las Vegas during autonomous-addled CES, the Delphi-Mobileye vehicle was the only one to venture into the realm of highway merges and exits. Those were signature moments in a drive in which the SQ5 cautiously proceeded along major arterials teeming with pedestrians, handled an oddball 120-degree turn at the bottom of a steep exit ramp, and navigated through the usual cast of human motorists paying scant attention to the rules of the road. Other than the few feet it took to go from the parking lot onto the road, the entire drive took place under fully autonomous control.
Twenty-one years ago, Delphi became the first automotive company of any kind to showcase its products at CES. In recent years, even as almost every major OEM and supplier has joined the company in transforming CES into another marquee destination on the auto-show circuit, Delphi has remained ahead of the pack.
Leaders in the burgeoning autonomous-vehicle industry have taken any number of approaches to developing their particular technologies, but they’d probably all agree that one essential component is a high-definition map. In collaborating with Mobileye, Delphi has taken what it believes is a crucial step toward an approach that reduces the role of a prominent sensor in depicting its surroundings.
Reducing the Role of Lidar
To date, most companies have used lidar to create these maps. In many ways, the whirring lidar sensors affixed to the roof racks of self-driving cars have become symbols of the fledgling autonomous era, and there’s little argument that these sensors provide high-quality data. But lidar has its disadvantages. For one thing, the sensors are expensive. In the current testing landscape, they can cost about $7500 per unit. Even as prices fall and suppliers like Velodyne project that solid-state lidars will cost as little as $50 per unit when high-volume production begins, OEMs and suppliers are still searching for alternatives.
“
“This is about getting to production in a realistic way, not with lidars all over the vehicle and equipment on top.”
—Glen DeVos, Delphi
”
That’s where Mobileye comes into the picture. The Israel-based company makes software for camera-based systems that are key components in the advanced driver-assistance systems installed in approximately 14 million vehicles worldwide. In August, it partnered with Delphi to build a turnkey self-driving system that prioritizes cameras and radar, rather than lidar, in its sensor hierarchy to figure out what the road ahead looks like.
“This is about getting to production in a realistic way, not with lidar [sensors] all over the vehicle and equipment on top,” said Glen DeVos, chief technology officer at Delphi. “This is a reasonable cost in a way that can be scaled up over time.”
Visualizing the Road Ahead
Working with Delphi, Mobileye engineers have developed what they call Road Experience Management (REM), a vision-based system that determines a vehicle’s location with a high degree of accuracy—within less than four inches—by measuring the distance between the vehicle and known landmarks it passes along the route.
Telemetry data from those measurements is then uploaded to the cloud, and Mobileye and Delphi get a precise idea of where multiple vehicles have traveled. Engineers can determine an average of where the past 1000 vehicles have traveled along a particular road and have a pretty good idea of the optimal route for its cars. Combined with real-time radar, lidar, and camera data, this average-path information can be a staple in delivering autonomous technology.
Committing enough vehicles to get anywhere close to 1000 passes on the same road would be an overwhelming task for even the most ambitious company, but what makes Mobileye unique is that it can harvest data from the 14 million cameras it already has on the road as part of driver-assist systems. Delphi and Mobileye see that network as a major advantage, especially since 20 automakers have agreed to make automated emergency braking a standard driver-assist feature in the United States by 2022.
“We’re harnessing the power of the crowd,” said Erez Dagan, Mobileye’s senior vice president for advanced development and strategy. “We can aggregate all this information into a highly dynamic, detailed map that gives highly automated vehicles all the tools to localize within it. And the bonus is, this map can serve as a foresight engine. You know if there’s a curve coming up or a highway exit that you need to take.”
Contrary to the high-definition maps produced by lidar, the data from Road Experience Management is significantly smaller in size, which in turn will cost less to send via over-the-air updates in near-real-time fashion. Mobileye spokesman Dan Galves said a map of the entire United States can be confined to approximately 64 gigabytes, and the data required for regular updates amounts to 10 kilobytes per kilometer of road.
“To me, the biggest thing that Mobileye has done is create a mapping data source that requires incredibly small data transfers,” said Mike Ramsey, research director at Gartner, a global technology consulting firm. “That’s the big advantage it poses. Whether you can actually eliminate or reduce the need for lidar, I think, is up for debate.”
Finished Product Two Years Out
Mobileye’s REM software, which runs on its proprietary EyeQ series of chips, is one component of Delphi’s overall autonomous strategy.
The company first made a splash in the autonomous industry in April 2015 by driving its trusty Audi SQ5 across the country, mostly under supervised autonomy. At the time, it needed its human safety drivers to negotiate the ramps on and off the highways. Two years later, the car is the same, but its autonomous underpinnings have changed.
In 2015, the supplier acquired Ottomatika, a Pittsburgh-based spinoff of Carnegie Mellon University that has software now governing path planning and control. In November 2016, Delphi partnered with Intel to provide the processing power for its software, which DeVos said will be about 10 times faster than current models. A fourth linchpin was added during CES, when Delphi announced it had acquired Movimento, a Detroit-area supplier that specializes in over-the-air update capabilities and cybersecurity.
Together, the Mobileye, Ottomatika, and Intel elements form the backbone of what Delphi calls its Centralized Sensing Localization and Planning automated driving system. In Las Vegas, the REM and Ottomatika portions operated on the test car, and Delphi said it intends to have the Intel chips integrated by late 2017 for testing and the entire system ready for production in 2019.
From there, Delphi intends to deploy the system as part of an ongoing pilot project involving autonomous vehicles in Singapore, in which the company anticipates using purpose-built vehicles both as self-driving taxis and for the on-demand shipment of cargo around cities. Should the timetable stay on track, that would give Delphi three years to test the system before putting it into service by its stated target date of 2022.
At that point, whatever the final sensor stack, DeVos still sees autonomous vehicles costing between $50,000 and $70,000, which is why he foresees autonomy reaching commercial markets long before most motorists can trade in their daily drivers for a highly automated vehicle. Commercial fleets, free to deploy vehicles at all hours, can pay back that cost much faster, and he sees the market for personally owned autonomy vehicles developing at a more incremental pace.
For now, DeVos is focused on the beginning of that transition, and he’s convinced the partnership with Mobileye has given Delphi a big head start.
“Really, this is our plan for getting to production with a program that helps us scale up,” he said. “What the partnership does is put vision and radar at the front of our sensor hierarchy, and from a vision perspective, accelerate our systems by many, many years.”
from Car and Driver BlogCar and Driver Blog http://ift.tt/2jBiaGR
via IFTTT
0 comments:
Post a Comment