Cars That Think iconCars That Think

photo of UAS Augsburg's driverless car on track during Formula Student Germany competition

Students Race Driverless Cars in Germany in Formula Student Competition

More than a dozen teams brought driverless cars to the Formula Student competition last week in Hockenheimring, Germany. It was the first event of its type, but many participants were diligent veterans of Formula Student Electric races and had tested their cars at different types of sites leading up to the main event. “We knew from the electric season that testing is really crucial,” says Manuel Dangel, vice-president of the Formula Student Driverless team at the Swiss Federal Institute of Technology (ETH) in Zurich. Then the rain started falling.

“We thought [our car] would basically fail,” Dangel says. While it had rained on one of their test days, their car’s main way of determining its own ground speed is an optical sensor optimized for dry ground. The team had not managed to complete a full ten-lap track drive in the rain.

Read More
A smartphone screen with text messages on it. The top of the screen reads "Travis"

Travis Kalanick's and Anthony Levandowski's Texts Tell the Tale of What Uber Knew About Waymo Tech

The epic court case between Waymo and Uber over self-driving car secrets took a tabloid turn last week, as Waymo’s lawyers filed a document containing approximately 400 text messages between Uber founder Travis Kalanick, and Anthony Levandowski, the engineer accused of taking thousands of files from Waymo to help build Uber’s lidar sensors.

Waymo expected the SMS messages, sent between February and December 2016, to reveal what Uber knew, and when the company knew it. Some emails do touch on technical matters—for example, one from Levandowski on 5 May saying that he was “driving to SF to meet with [Uber’s] laser guy and guide the team.”

However, Waymo’s lawyers say that there are “significant and inexplicable gaps” in the text messages, including none at all before February 13, even though the two had certainly met before. Despite this, the texts provide a rich insight into the men’s relationship, and into Uber’s plans for (and worries about) its self-driving car technology.

The Otto Acquisition

Just two weeks after Levandowski quit Google, Kalanick was already visiting the engineer’s new self-driving truck start-up, Otto. Kalanick was planning to buy Otto almost immediately but that fast pace came with issues:

2/13/2016 Kalanick: Good hangin

2/13/2016 Levandowski: It was awesome. Lots more to come. We ended up wrapping truck testing at 2.30

2/13/2016 Levandowski: We had a close call but no contact with anyone or anything

This appears to be a reference to a failure of the self-driving technology that nearly resulted in an accident. Presumably, the testing was happening at a test track, as the California Department of Motor Vehicles (DMV) still does not allow the testing of autonomous commercial vehicles.

4/6/2016 Levandowski: Basically I’d like the freedom to move as needed on the acquisition (and take advice/guidance) but if I can close them within the range we agreed you guys are happy.

4/6/2016 Kalanick: I am super down to make sure this [is] quick lightweight and straight forward for you guys

At this point, Otto was being run from Levandowski’s home in Palo Alto. In early April, the DMV launched an investigation that had Levandowski and Kalanick worried.

4/22/2016 Kalanick: How did they find out?

4/22/2016 Levandowski: Trying to dig in, likely city of Palo Alto.

4/22/2016 Levandowski: Just wrapped with the DMV, it was the city of Palo Alto freaking out about AV trucks testing and were asked to investigate. The guys were happy with our answers and were [sic] in the clear.

Levandowski and Kalanick’s Relationship

The two men quickly formed a strong bond, but there are challenges—and advantages—when your new best friend is the CEO of the world’s largest start-up.

3/29/2016 Levandowski: I am at the secret side door, no rush

7/23/2016 Kalanick: You hungry? .. Can get some Uber Eats steak and eggs.

At these meetings, often late at night, Levandowski would explain the mysteries of self-driving technology to the Uber founder.

4/8/2016 Kalanick: Where you teach me in depth about an autonomy topic

4/8/2016 Levandowski: Yes, we should of done it. We did a bit on lasers before but need to go deep on all the topics.

In return, Kalanick dispensed management advice, such as this just before the Otto acquisition was announced:

8/12/2016 Kalanick: Three principles

8/12/2016 Kalanick: 1) don’t tell anyone about the deal before it happens, ESPECIALLY someone you're about to fire 2) firing fast is a cultural imperative you don't want to break except in the most extreme situations 3) get creative

Both men shared the same ambition:

9/19/2016 Levandowski: We’re going to take over the world

9/19/2016 Levandowski: One robot at a time

10/7/2016 Kalanick: Down to hang this eve and mastermind some shit

Uber Really Wanted to Partner With Google

Uber’s rivalry with Alphabet’s self-driving subsidiary Waymo is a recent thing. An earlier court filing contained an email from 2015 that showed Kalanick and Google founder Larry Page were exploring a partnership on self-driving technology. The new text messages suggest that this was still a hope over a year later.

6/13/2016 Kalanick: Just got word from Drummond that g-CO is out

6/13/2016 Levandowski: Wow, at least now we know it's a zero sum game

David Drummond is Alphabet’s chief legal officer and was a board member of Uber until August 2016. He was the main channel of communication between the companies. “G-co” could refer to cooperation or forming a company with Google, the lack of which cemented the conflict between the two, and ultimately pushed Drummond off the board.

Uber Saw Tesla as a Huge Competitor

While Uber followed Google’s cars closely, it was Tesla and Elon Musk that the duo discussed most frequently.

9/14/2016 Levandowski: Tesla crash in January … implies Elon is lying about millions of miles without incident. We should have LDP on Tesla just to catch all the crashes that are going on.

9/22/2016: We’ve got to start calling Elon on his shit. I'm not on social media but let's start "faketesla" and start give physics lessons about stupid shit Elon says like [saying his cars don’t need lidar]

In late October, the two exchanged a flurry of tests about Musk’s announcement that all Teslas would come with all the hardware necessary for full self-driving, sometimes called Level 5.

10/20/2016 Levandowski: Elon is going to make going to [self driving] not as big of a scary thing for the public... which should be good

10/20/2016 Kalanick: Got to get software runnin

10/20/2016 Levandowski: Amen

10/20/2016 Kalanick: What do you think chances are he has Level 5 in 20% of a given city?

10/20/2016 Levandowski: For easy city

10/20/2016 Levandowski: He's trippin' but might/will blame regulatory as to why it's not available

Did Uber’s Cars Have Real Problems in San Francisco?

In December 2016, Uber launched a self-driving taxi service in San Francisco, without obtaining permission from the DMV. The program lasted only a week, and was dogged by reports of Uber’s 16 cars running red lights. A single text from Levandowski to Kalanick, two days before Uber’s cars had their registrations revoked by the DMV, refers to the issue:

12/19/2016 Levandowski: Quick update on that special intersection in SF, we taped 6 red car violations within 2 hours

A source close to Uber’s operations says its engineers watched the intersection where Uber’s cars were said to have run the red light, and that this text refers to them recording a number of normal, human-operated vehicles also breaking the law. Uber has never officially admitted that its software was to blame.

This post was corrected on 15 August 2017 to fix the context of a 12/19/2016 message and on 16 August 2017 to correct the October text message dates.

A black satellite in a white room appears to have a small nose and prominent ears

Sapcorda Plans Centimeter-Scale GPS for Europe

Germany’s Bosch and Geo++, U-blox of Switzerland, and Japan’s Mitsubishi Electric announced the establishment of Sapcorda Services last Tuesday, a joint venture to provide global navigation satellite system (GNSS) positioning services of centimeter-level accuracy via satellite transmission, mobile cellular technology, and the Internet.

Read More
A small off-white plastic box with two protruding black antennas

E-ZPass Could Kickstart Smart Cities

Everyone likes the idea of a smart city. Traffic lights would automatically adjust to optimize traffic flow, you could find parking spaces without circling for hours, and enforcing speed limits wouldn’t require a cop on every corner.

The problem is that technologies currently underpinning these systems are often expensive, unreliable, or both. Cameras need pricey computer vision systems to know what they are looking at, and require cleaning, while sensors embedded in the road are too expensive to monitor all but the priciest parking spots. And radars are good at measuring speeds, but poor at specifying individual vehicles.

Now researchers at MIT have designed a smart city system that leverages the windshield tags used to smooth drivers’ passage onto toll roads and bridges. E-ZPass tags in the eastern United States—and similar systems nationwide and around the world—are radio-frequency transponders that transmit a unique signal when queried at a certain frequency. That response is typically picked up by a reader mounted on a gantry over the highway, and the driver’s credit card is automatically charged.

However, Dina Katabi and Omid Abari at MIT realized that anyone could prod a transponder to emit its signal, simply by asking nicely. They have developed a small solar-powered unit called Caroake that can measure the position and speed of up to 20 nearby vehicles that are equipped with E-ZPass transponders. Each Caroake unit currently costs around $100, although that figure would drop significantly if the units where produced in volume, say the researchers.

By putting Caroakes on every urban streetlight, the researchers envisage tracking the occupancy of every parking space, counting how many cars are waiting at every traffic light, and perhaps even catching every speeding vehicle. They are about to deploy six readers on public streets in Cambridge, Massachusetts to learn local traffic and parking patterns.

“There is huge interest both from the city and from E-ZPass, who see this as a mechanism both to expand the functionality of a device that already exists in the vast majority of cars, and to make the city smarter,” says Katabi.

But the pilot involves much more than simply putting normal E-ZPass readers on poles. The highway readers use highly directional antennas to ensure they are querying a single car at a time as it passes beneath them. That technology is both expensive and of limited use in a city, where you want a more holistic view of what is happening along each street.

Instead, the Caroake device activates all E-ZPass tags within a 30-meter range. “In a city, 20 or 30 cars might respond simultaneously, the signals collide and you get a real mess,” says Katabi. Luckily, although all E-ZPass tags are meant to broadcast at 915 megahertz, in practice individual units tend to stray a little to either side of that frequency. The MIT system uses this so-called carrier frequency offset to separate the signals from one another.

Localizing each tag is another challenge. A pair of antennas in the system calculate the signal’s angle of arrival. For stationary cars, that is enough information to locate them in a parking spot. To track moving cars, two or more neighboring readers can work together to triangulate its position. Caroake units form an ad hoc mesh network with each other, and use a Wi-Fi or cellular LTE gateway to push data up to smart city servers.

In tests on MIT’s campus, the Caroake network could determine speed to within 8 percent (about 1.6 to 4.8 kilometers per hour) and location to within 4 degrees, which was accurate enough to place cars in individual parking spots.

Another worry for the researchers was that constantly querying the E-ZPass tags might run down their sealed, non-rechargeable lithium batteries. In normal use, where they might be queried just a few times a day, E-ZPass tags can last a decade. In a smart city, the same tag might be activated a hundred times daily, or more. “We ran experiments and found out that even if you query the transponders every second, their batteries will still run for around 10 years,” says Abari.

But the biggest concern for many drivers is likely to be the privacy and security issues that come from being tracked throughout a city. In 2013, the New York Civil Liberties Union found that city’s Department of Transportation had quietly deployed dozens of standard E-ZPass readers to measure traffic volumes and congestion. The Union called the department’s privacy policies at the time “vague and barebones.”

In their pilot tests, the MIT researchers will measure only the carrier frequency offsets of the transponders and not decode the identifying data bits. “We do not believe that [these] values can be mapped to the owners or used to infer any private information about them,” they wrote.

Of course, if E-ZPass tags were to be used for cash-free parking, they would have to link individual cars to their owners. Dina Katabi thinks that far from infringing people’s privacy, Caroake devices could enable new, convenient business models: “It could be used to pay for other services. For example, I could drive to a restaurant and be charged immediately, or get access to certain roads at certain times, like congestion charging.”

For those services to be available to everyone—and for the smart parking and traffic systems to be most effective—every car will need an active transponder. Adoption rates are creeping up as states like Massachusetts and Pennsylvania sunset cash toll booths in favour of E-ZPass, but they are still at only 80 to 90 percent.

Perhaps a bigger problem is that anyone worried about Caroake’s Big Brother implications (or who just likes to speed) could temporarily pop their tag in the metallic “read prevention” bag that E-ZPass provides with every unit. After all, smart cities will be home to plenty of smart people.

cutaway 2015 nissan leaf

Car Battery Breakthrough Claimed in Japan

Mitsubishi and GS Yuasa say they’ve developed a lithium-ion battery that will give electric cars twice the range that today's designs do, for the same cost.  The report, in yesterday's Nikkei Business Daily, adds that the battery will go into mass production in 2020.

Such a battery would at the very least assuage range anxiety among potential car buyers. It could even allow manufacturers to shrink the battery, and thus cut the car’s weight, energy usage, purchase price, and operating cost.

It’s the latest in a string of announcements that imply progress at least as fast as the e-car’s fondest proponents have claimed. Last week, a consortium of German companies revealed plans to build a lithium-ion battery gigafactory to rival Telsa's, with production slated to begin in 2019. And in June, Toyota—a company that had long talked up fuel cells as the natural successor to batteries—apparently changed direction by touting a new lithium-ion battery design using a solid electrolyte. Because this solid-state design would be less sensitive to heat, engineers should be able to pack cells more closely, cutting weight. The design might also allow for faster charging. Also in June, Volvo said its next-gen models would all employ electric drive, either alone or hybridized with a gasoline engine.
And in the most comprehensive report yet, Bloomberg New Energy Finance says that electric cars will reach cost-parity with their fossil-fuel kin by 2026. Here’s Bloomberg’s projection:

Electric cars approach price-parity with conventional ones

But as Matthew Eisler pointed out in this space last month, the projection is based on the cost per cell, leaving out the highly pertinent metric of battery pack lifetime. That may not be much of a problem for your cell phone’s battery, given the short lifetime of the phone. But nobody knows yet whether the new automotive lithium-ion batteries will need to be replaced before the car wears out.

Then again, it may not matter all that much. Range anxiety may turn out to be more a theoretical than a practical problem, particularly now that electric cars are packing serious kilowatt-hourage.

A couple of years ago, scientists at the Lawrence Berkeley National Laboratory found that even when a car’s batteries have lost 30 percent of their storage capacity, most people could make the trips they normally make. And that study was based on the 2015 Nissan Leaf, which then had a range of just 135 kilometers (83 miles). The new Chevrolet Bolt can go nearly three times as far on a single charge. 

Minor changes to street sign graphics can fool machine learning algorithms into thinking the signs say something completely different

Slight Street Sign Modifications Can Completely Fool Machine Learning Algorithms

It's very difficult, if not impossible, for us humans to understand how robots see the world. Their cameras work like our eyes do, but the space between the image that a camera captures and actionable information about that image is filled with a black box of machine learning algorithms that are trying to translate patterns of features into something that they're familiar with. Training these algorithms usually involves showing them a set of different pictures of something (like a stop sign), and then seeing if they can extract enough common features from those pictures to reliably identify stop signs that aren’t in their training set.

This works pretty well, but the common features that machine learning algorithms come up with generally are not “red octagons with the letters S-T-O-P on them.” Rather, they're looking features that all stop signs share, but would not be in the least bit comprehensible to a human looking at them. If this seems hard to visualize, that's because it reflects a fundamental disconnect between the way our brains and artificial neural networks interpret the world.

The upshot here is that slight alterations to an image that are invisible to humans can result in wildly different (and sometimes bizarre) interpretations from a machine learning algorithm. These "adversarial images" have generally required relatively complex analysis and image manipulation, but a group of researchers from the University of Washington, the University of Michigan, Stony Brook University, and the University of California Berkeley have just published a paper showing that it's also possible to trick visual classification algorithms by making slight alterations in the physical world. A little bit of spray paint or some stickers on a stop sign were able to fool a deep neural network-based classifier into thinking it was looking at a speed limit sign 100 percent of the time.

Read More
interior of tesla model 3

Tesla Model 3 to Driver: Look Me in the Eye

Driver-facing cameras aren’t a new idea: Cadillac’s CTS Super Cruise uses an infrared camera to monitor your eyes for signs of inattention. So does Audi’s new A8, billed as the world’s first car capable of Level 3 autonomy.

So you’d expect Tesla to put one in its new Model 3, given that it’s supposed to have all the hardware it’ll need for both Level 3 and Level 4 autonomy. (Level 3 requires the driver to be prepared to take control, but Level 4 does not.)

The funny thing, though, is that the Model 3’s driver-oriented camera isn’t turned on. In fact, its very presence wasn’t mentioned by the company or noticed by the auto writers who first drove the car. So a tip of the hat goes to Fred Lambert, who confirmed the camera’s existence yesterday, in Electrotek.

According to Elon Musk, all that’s needed for full-bore self-driving power is an over-the-air software download. That, however, won’t come until regulators allow it. So don’t hold your breath. Same goes for actual, on-the-road autonomy for the Audi A8.

Another funny thing is the deliberate inclusion of what you might call sleeper hardware. You’re selling customers something that is of no use now, but will be—you say—later on. Sleeperware is something you don’t often seen in techland because of the fast product turnover, but even cars practically never have it. 

My first encounter with sleeperware dates back to 1993, when my office upgraded to Macintosh computers that each had a little indentation right above the screen. 

“What in the world is that,” I asked a colleague. “It’s a built-in microphone—they think computers will be used as communications devices,” he responded. Hah, I thought.

But it happened, and fast—within the working life of those 1993-era desktop machines. So maybe the Model 3 you’re driving will one day drive you.

Engineers look at self-driving test vehicles at a track day

Carmakers Take to the (Self-Driving) Test Track

Researchers from academia and industry took rides in experimental cars at a public test-track event in Teesdorf, Austria, last week, but the main draw may have been the other attendees.

The event gave smaller companies a chance to try out driverless technology on a shared large-scale test track. Formal vehicle testing on closed tracks can cost up to £1000 (US $1320) a day. “We thought we could do something that was a bit different: combine the opportunity for small companies and university teams,” says event organizer Alex Lawrence-Berkeley, of Sense Media Group in London, England.

Read More
Christina Lampe-Onnerud, CEO of Cadenza Innovation

Christina Lampe-Onnerud Has A New Battery Design

Most attempts to improve lithium-ion batteries involve changes in the chemistry or components of a cell. But you can also work wonders on the larger scale of the battery pack—the array of cells that power big machines, like electric cars.

That’s the tack taken by Cadenza Innovation, a Connecticut firm started up by Christina Lampe-Onnerud, a Swedish battery maven we profiled back when electric automobiles were just getting going (“The Lady and the Li-ion,” March 2008). And because the strategy, which she calls a supercell architecture, allows for the packaging of different cell shapes and sizes, it means serving existing battery companies rather than supplanting them.

She declines to reveal which manufacturers Cadenza is negotiating with.

Lampe-Onnerud’s earlier initiatives centered on the cell itself, but that required her to play the sharp-elbowed game of mass manufacturing. That game is far harder to get into now that megacorporations have started building battery gigafactories.

“We simplify the architecture and manufacture of the batteries, eliminating quite a few parts,” she says. “That means there are more energy-carrying components and fewer surrounding components.” You can get from 30 to 100 percent more watt-hours per kilogram of battery pack, depending on the kind of cell being used and the number of cells in the array.

Besides increasing energy density, the idea also cuts cost. One electrical connector can yoke together 20 or 30 cells, for instance. And, by offering a way to keep the familiar “jelly roll” cylindrical cells separate from one another, in a heat-resistance plastic frame resembling an egg crate, Cadenza makes it safer to be around all that close-packed energy.

Other safety devices enter the mix as well, like a vent, a pressure dome, and a mechanical means of connecting and disconnecting cells. Of course, the tradeoff of energy density and safety goes in both directions: A customer can choose to pack even more cells into a battery while maintaining existing safety standards.

“You can say we’re packaging 20 years’ worth of best practice into a Lego block,” she laughs. “And it’s coming out when one of the best performaners, Samsung, overstepped what it should have done with its Galaxy phone.” 

Volvo Cars CEO Hakan Samuelsson standing in front of a blue Volvo car and a white Volvo car

Will Volvo Really Kill the Gasoline Engine?

At first glance, Volvo’s announcement earlier this month seems radical: Every new vehicle the Swedish automaker launches from 2019 will have an electric motor. By 2025, Volvo hopes to sell one million electrified cars in a plan it claims will herald the end of the era of automobiles equipped only with an internal combustion engine.

These declarations elicited the usual mixture of enthusiasm and skepticism. After all, automakers have promised the moon many times before, only to fall short. Can this storied firm cause a paradigm shift in propulsion technology?

If history is a guide, probably not. For at least the next decade or so, Volvo will continue to produce purely gasoline-powered cars, especially its profitable SUVs. Nor will all of its electric cars be all-electric; Volvo will also make plug-in and conventional hybrid electrics.

Read More

Cars That Think

IEEE Spectrum’s blog about the sensors, software, and systems that are making cars smarter, more entertaining, and ultimately, autonomous.
Contact us:

Philip E. Ross
New York City
Willie D. Jones
New York City
Senior Writer
Evan Ackerman
Washington, D.C.
Mark Harris

Newsletter Sign Up

Sign up for the Cars That Think newsletter and get biweekly updates, all delivered directly to your inbox.

Load More