What Happened When We Took the SCiO Food Analyzer Grocery Shopping

Who needs infrared spectrometers in their phones? People who hate buying tasteless produce or mystery cheese
Photo: Tekla Perry

I’m at a Whole Foods in Palo Alto with Dror Sharon, cofounder and CEO of Consumer Physics, based in San Francisco and Israel. Sharon is holding his smartphone and a tiny handheld device he calls SCiO, which is about the size of a TicTac box. We are browsing around the produce department, checking out the Brix level of various items. The Brix number represents the sugar content of a solution and, for fruits, is an indicator of whether or not a particular fruit has much flavor. The tomatoes, according to the SCiO’s accompanying smartphone app, are horrible; not a big surprise in March. The apples are mixed, there is only one variety Sharon would buy right now. The mangos, he proclaims, are just perfect, and contemplates filling a bag before we go.

We move onto the dairy case, where the labels of cellophane-wrapped cheeses provided only price and name. Sharon’s smartphone app popped up all sorts of additional information as he pointed the SCiO gadget at different chunks (still in their wrapping), including fat content, calories per gram, and protein content.

On the way to Whole Foods, we stopped outside a restaurant where two women were having brunch, and asked them if we could scan their food before they ate it. Sharon told them the strawberries would be excellent (they women agreed they were), but the whipped cream would be abnormally sweet, there was so much sugar in it wasn’t recognizable as dairy (it was).

It was all pretty magical, pointing a gadget at food and getting an instant analysis. To be fair, I can’t verify the accuracy of what I was seeing on the screen; I didn’t take the fruits and cheeses back to a laboratory to confirm the analysis using more traditional technology. But it certainly seemed real, real enough that I would be pretty excited to have this kind of technology built into my smart phone, given I have my phone out anyway when I’m grocery shopping to scan shelf tags in order to download coupons. And Sharon promises it is indeed coming into phones—as soon as the third quarter of this year in China, fourth quarter in the United States.

Here’s how SCiO works—and why it exists.

The gadget uses standard infrared spectroscopy; it measures the absorption of infrared light. It may not be as accurate as a benchtop spectrometer used in a laboratory environment, but Sharon says it makes up for this with its algorithms. The user starts out by simplifying the problem a bit by identifying the category of the item to be examined—it’s not “What fruit is this,” but, “This is an apple, is it any good?” Consumer Physics’ cloud-based software then taps into its knowledge base, for an apple, it defines “good” as “sweet” (hence the Brix measurement), and considers an apple’s typical range of sweetness based on thousands of scans. A graphic on the phone then places the apple on a quality range.

Besides having data on most fruits and vegetables, the system also knows about dairy products; for those, it provides information on calories and fat content. And it knows about the cocoa content of chocolate, the amount of alcohol in drinks, and the protein, fat, and calories in raw fish, poultry, beef, and pork. And while, to date, the focus has been on food, Sharon stresses that the technology works with all sorts of materials. The company has started holding workshops for people who want to develop their own databases.

Sharon had been wanting this kind of gadget for a long time before he finally set out to build one. He grew up on a farm in Israel; he was used to eating produce that hadn’t been shipped further than across the property. So, when he moved to Massachusetts for business school at MIT (his bachelor’s degree is in electrical engineering), he was surprised by just how tasteless he found the produce at local groceries. “The food just didn’t taste the same. And when I saw that I was buying grapes from Chile, I was sure something was not right about them.”

He decided that he should get himself something to determine whether or not the food in the stores was any good before he bought it, so he logged onto Amazon and searched for such a gadget. He didn’t find one. Disappointed, he resigned himself to occasionally buying tasteless produce or traveling 30 miles to a grocer he discovered that he could trust.

But about five years later, in 2010, after a few years working in the U.S. and then moving back to Israel, he came back to the idea. There ought to be a scanner that could give you useful information about the food you are about to buy, he insisted. He teamed up with Damian Goldring, a friend from his undergraduate days with a PhD in silicon photonics, and the two started investigating sensing technologies that, potentially, could be built into a phone. They landed on infrared spectrometry, and, in 2011, started Consumer Physics. In mid-2012, they rented one of those expensive, luggable, commercial spectrometers for a day and demonstrated to a large cellular service provider that the technology could be used to analyze food, doing a demo on chocolate mixtures that looked the same, but had different substances mixed in, like regular butter and peanut butter. “We’re going to put this into a phone,” Sharon said. (The company didn’t fund them.)

Sharon and Goldring may not have convinced that company, but they had convinced themselves, and began working on the technology, first on their own dime, and then with a little money from angel investors and crowd-sourced funding from OurCrowd. In early 2014, they were convinced enough that they could deliver the technology as a small Bluetooth peripheral—not inside a phone quite yet, but pretty close—to launch a Kickstarter campaign, pitching a $200 portable infrared spectrometer. Some 13,000 people signed up, ponying up about $2.7 million.

Things from Kickstarter funding to shipped product were not exactly smooth sailing. Come September of 2016, we reported that only 5000 of the Kickstarter backers had received products, far later than originally estimated, and many of the remaining backers were angry. To make things worse, the backers could no longer communicate with the company via Kickstarter, the page had been taken down in a trademark dispute over the name “SCiO”.

What happened? Sharon says the delays were due to manufacturing challenges, as well as a redesign to improve sensitivity, resistance to ambient light, and penetration depth. And the company has now fulfilled almost all of its Kickstarter orders, with the exception of customers who haven’t yet provided shipping addresses, have unique shipping requirements, or are choosing to wait for a Special Edition version of the gadget—that’s fewer than 10 percent of the backers, Sharon says.

But while the Kickstarter rollout was more than normally bumpy, the company’s efforts to get venture funding have born, well, fruit. After picking up some funding from angel investors and people using crowdfunding platform OurCrowd, Consumer Physics closed a round of venture investment led by Khosla Ventures. To date, Sharon said, funding totals over $25 million.

The company also lined up some critical partnerships: with Analog Devices, which worked with the company to reduce the size of the sensor package into something that will easily fit into smartphones and is manufacturing this version of the device; and with Chinese phone manufacturer Changhong, which will be incorporating the technology in the Changhong H2 smartphone starting in China in the third quarter of this year and in the U.S. towards the end of 2017. Consumers in China, Sharon points out, are particularly interested in checking food safety, given the history of problems with the food supply. Sharon hopes other smartphone manufacturers will follow, turning using a phone to scan food as common a practice as using one to photograph food.

Consumer Physics now has about 100 employees, with corporate offices in San Francisco, a sales team based in the Midwestern United States, and a development team in Israel. Dozens of people are scanning food 24/7, Sharon said, to increase the kinds of food that can be analyzed as well as the accuracy of the analysis.

While the initial applications surround food, Sharon says that the technology is not just for checking out food freshness and nutritional information; it’s good at analyzing body fat, and distinguishing real pharmaceuticals from their fake counterparts.  “We’ve done a demo that distinguishes real Viagra from fake Viagra,” says Sharon. “That’s the most commonly counterfeited drug.”

Consumer Physics has, to date, shipped more than 3000 developer kits, and is hoping some interesting consumer applications will emerge. One such in the works by French company Terallion, Sharon said, is a kitchen scale, intended for diabetics, that can use SCiO’s analysis to allow it to give users accurate information about protein and carbohydrate content of the food they are about to eat. The company is also working directly with industrial partners, in particular, with those working to develop tools for digital agriculture.

The Nobelists and Their Molecular Machines

Development is incremental in molecular nanotechnology, but it is coming along slowly
Illustration: CiQUS/Royal Society of Chemistry

While the prospects of molecular nanotechnology—the catch-all term for molecular manufacturing in which nanoscale machines are programmed to build macroscale objects from the bottom up—has remained mostly in the realm of science fiction, the awarding of last year’s Nobel Prize in chemistry to a trio of scientists who pioneered the development of nanomachines has buoyed hope that at least we should begin to see more research in the field.

More of this research is already trickling in since the Nobel Prize announcement. Two teams of researchers from the University of Santiago de Compostela (USC) in Spain have cited this most recent Nobel Prize as a context for their work in developing self-assembling materials based on peptides (compounds consisting of two or more amino acids linked together in a chain) that can stack themselves on top of each other to form nanotubes.

In two studies—one described in the Journal of the American Chemical Society (JACS) and the other in the journal Nanoscaleseparate teams of researchers under the supervision of USC professors Juan R. Granja and Manuel Amorín observed the synthesis of self-assembling materials based on cyclic peptides that, under the right conditions, can stack, or self-assemble, into very particular shapes.

In the research upon which the JACS paper was based, the peptides formed into a capsule that can selectively recognize a specific type of ligand (a substance that forms a biomolecule) based on its shape, size and functional group. Depending on what the ligand is, the peptide capsule will capture it.

While being able to recognize these three characteristics of a ligand is not so usual, the capsule has three different reversible bonds in its formation. These types of bonds can be used to trigger the liberation of the confined molecule when conditions change. In addition, the amino acid used in the capsule can be modified without affecting the encapsulation properties, opening up the opportunity to target the capsule to specific cellular receptors. Also the peptide character could provide some biocompatibility. All of this points to a potential use as some kind of drug delivery tool.

In the other research, described in Nanoscale, the researchers were able to shape a carbon nanotube into Venturi tubes that are used for constraining fluid long enough to be measured. The researchers were able to achieve this shape by covalently linking two cyclic peptides of different size.

While the formation of carbon nanotubes is obviously not dependent upon peptides operating as nanoscale machines, this particular shape is unique.

“The synthesis of carbon nanotubes with this shape has not yet been described,” explained Granja in an e-mail interview with IEEE Spectrum. “It is still quite complex to control all the structural properties of the carbon nanotubes (diameter, length, and the way the graphene sheet is wrapped).”

Granja points out that that never before have carbon nanotubes been prepared by the self-assembling of smaller components. “In fact, this is one of the advantages of the peptide nanotubes,” Granja says. “This method allows the precise control of the nanotube diameter,” 

While capsules that can identify and contain molecules and serve as nanoscale Venturi tubes sound like they are a bit more applications-oriented than nanocars, it all still seems pretty far removed from the day when engineers are able to watch as nanomachines build a macroscale car from the bottom up. So where are we on that timeline, especialy now that the Nobel committee has shined a light on molecular manufacturing?

Granja does see USC’s most recent work providing a small contribution to the development of the future components of complex molecular machines, but concedes that we are still far away from the development of programmable machines that can carry out specific complex functions programmed by the inventors.

Nonetheless, there are some clear landmarks lying ahead that we should be keeping an eye on as signs of progress, according to Granja.

“The design of artificial ribosomes to synthesize any type of complex biopolymer with similar efficiency and activity of the natural ribosome should be one of the big goals,” said Granja. “The construction of synthetic ion channel pumps must be another future goal. The final dream should be that the molecular machines should be able to generate under specific conditions from smaller components to carry out the work.”

While these incremental steps remain pretty far from the self-repairing machines envisioned by Eric Drexler in his book “Engines of Creation: The Coming Era of Nanotechnology”, Granja believes these steps are what will eventually lead us there. One of the things that seems to be increasingly clear is that nanomachines are looking to be more biological than machine-like bots, just as Richard Jones of the University of Sheffield suggested nearly a decade ago here on the pages IEEE Spectrum .

Along this bio-bot line of research, Granja believes that a lot of excellent work has been carried out since the pioneering work of professors Bern Feringa, Jean-Pierre Sauvage, and Sir J. Fraser Stoddart, last years Nobel Laureates.

While Granja believes his group’s most recent contributions to the field are modest, he does see the Venturi-like nanotubes research contributing ultimately to the development of ion channels through molecular manufacturing. To date, these ion channels, which can be used to transmit electrical signals, have only been fabricated from the top down, often using semiconductor manufacturing techniques. If they could develop a way to make them from the bottom up, using molecular manufacturing, it could lead to a new era in which these devices could be fabricated more precisely and deliver better performance.

Fractus Antennas Pitches New “Antenna-less” Smartphone Technology

Instead of a dedicated antenna, the company's approach radiates radio-frequency signals from the ground plane
Photo: Amy Nordrum
Employees at Fractus Antennas in Barcelona use this anechoic chamber to test new antenna designs. Spikes along the walls absorb stray signals while sensors around the middle ring measure signal performance.

A tiny company based in Barcelona is promoting a new technology that it hopes can revolutionize smartphone antennas—by removing them altogether.

The job of any smartphone antenna is to radiate a radio-frequency signal generated by the phone’s transmitter out to the nearest cellular base station or Wi-Fi router. Now, Fractus Antennas wants to replace that antenna with a much smaller component called an antenna booster—a tiny lightweight cube made of a metal frame and FR-4 epoxy, the same material used in printed circuit boards.

The company says it can use this booster, along with some modifications to the smartphone’s circuitry, to radiate RF signals exclusively from a device’s ground plane—with no dedicated antenna to speak of. According to the company, this approach can deliver performance comparable to today’s smartphone antennas, at a lower cost for manufacturers.

During this year’s Mobile World Congress, the smartphone industry’s largest trade show, Carles Puente, Fractus Antennas’ cofounder and vice-president for innovation, quietly wandered the exhibit halls and handed out samples of this antenna booster from his satchel to any smartphone maker who might be able to use it.  

Back at his office near Barcelona, Puente compared it to a few antennas found in smartphones over the past decade. He pulled several devices from storage that were left over from the time his first company, Fractus, which also specializes in antennas, sued 10 manufacturers for patent infringement. He and his staff broke open more than 600 smartphones to build their case, so they’ve seen more than their share of internal antennas.  

To explain how the booster technology from his new company, Fractus Antennas, works, he first showed me antennas from a 2008 Blackberry Pearl and a Pantech C740 from the same year. With the casing removed, it was easy to see that both models had what looked like a tangle of metallic lines toward the top of the device. Those squiggly patterns were mounted to plastic structures that gave the antennas a shape designed to help them radiate energy most effectively.   

Puente told me these antennas are all inspired by fractals, a type of design in which similar patterns repeat themselves at various sizes. Fractals are naturally found in broccoli stalks and tree branches. With a fractal-based design, smartphone manufacturers can use all or just part of an antenna to provide service across many frequency bands.

For example, one of the longest wavelengths that smartphones must support is for the 698-megahertz frequency, where waves measure 430 millimeters long. And because the size of a radio wave corresponds to the size of the antenna needed to transmit it, longer wavelengths require larger antennas.

A typical smartphone antenna might only be 40- to 60-mm long, so transmitting waves that long requires the antenna’s entire surface. Since smartphones must also provide service across five or six other frequencies, smaller chunks of the same antenna are used to transmit those shorter wavelengths.

Fractus Antennas is a spinoff of Puente’s first company, which patented the use of fractal-based antennas in smartphones (and eventually filed that 2009 lawsuit for patent infringement). For many years, those were the dominant type of antenna found in smartphones.

More recently, manufacturers have moved away from fractal-based antennas and simply placed a metal band along the top of the smartphone to serve as an antenna. But one drawback of these metal bands is that they can’t easily support multiple frequencies at the same time on their own.

Manufacturers must add another part—an active tuner—to generate signals at the frequencies required for carriers around the globe. Still, this tuner is best at providing service at either one band or another, rather than over multiple bands at once.

Meanwhile, the industry is moving toward interband carrier aggregation, in which a device combines spectrum from several frequency bands to build a channel with more bandwidth than would otherwise be available. If metal band antennas can’t simultaneously provide service across bands, they may not be very useful as carrier aggregation becomes more popular.

This is where Fractus Antennas’ new, “antenna-less” smartphone technology comes in. Instead of relying on a dedicated antenna to radiate an RF signal, a handset would radiate the signal directly from the ground plane, which is the copper layer that underlies the phone’s printed circuit board. To do this, the phone’s manufacturer would replace the antenna with Fractus’s  mXTEND Antenna Booster a small device roughly one-tenth the size of a traditional antenna. 

It works like this: Once the transmitter generates a signal, it travels through the matching network, which is a part of the smartphone that acts like a tuner to support service at various frequencies. From there, it travels to both the booster and the ground plane. The booster is a passive device that does not radiate at all. Rather, it temporarily stores the signal it receives and repeatedly bounces it over to the ground plane, which radiates it out.

Already, today’s smartphones use the ground plane to radiate a portion of the signal that smartphones produce. To prevent interference, their circuit boards incorporate shields to protect elements that may be vulnerable. However, the Fractus Antennas concept takes this to the next level by using the ground plane to produce all of the radiation that is broadcast to the cell tower or Wi-Fi router. “Instead of having an antenna that radiates inside the phone, the phone itself is radiating,” Puente says.

The antenna booster does require a slightly more complicated matching network than usual. Puente says the matching network of a phone with an antenna booster would include six or seven components rather than the one or two found in a smartphone today—and the network must be redesigned for each model.  

Fractus Antennas is now selling several versions of its antenna booster, which can support cellular communications across 12 frequency bands (from 698 MHz to 2690 MHz) and can also be adapted for Wi-Fi and Bluetooth. The company launched in 2015, and its first sales came in 2016.

So far, Fractus Antennas has sold hundreds of thousands of units to a dozen clients who are using them to track fleets of trucks and outfit sensors for smart metering, among other things. Right now, it costs Fractus Antennas about US $1 to produce each unit, but Puente expects they could reduce that cost considerably by producing higher volumes.

The company is not claiming that the booster improves performance; in the company’s tests, it has shown its performance to be similar to that of today’s smartphones. Puente believes its main selling point will be the money that smartphone makers can save by never having to design and manufacture their own antennas again.

If manufacturers sign on , Puente predicts that it may be 2018 before Fractus Antennas’ technology is available in a smartphone. That, he says, is thanks to the devices’ long development cycles. The 15-person company is an underdog in the industry that generates annual revenues of more than US $400 billion worldwide.  But as Puente learned from his first company, a few strong patents can take a company far.

One Small Step for a Paraplegic, One Big Step Toward Reversing Paralysis

A clinical trial in Switzerland is testing a spinal implant to help paralyzed people walk again
Photo: Hillary Sanctuary / EPFL
A man who was partially paralyzed by a spinal cord injury is testing a spinal implant to help him walk again.

In a hospital in Switzerland, permanently paralyzed people are now learning to walk again with the help of stimulating electrodes implanted in their spines. For Grégoire Courtine, professor of neuroprosthetics at the Swiss Federal Institute of Technology Lausanne (EPFL), this day has been a long time coming. “It took us 15 years to get from paralyzed rats to the first steps in humans,” he says. “Maybe in 10 more years, our technology will be ready for the clinic.”

Courtine has made it his mission to reverse paralysis. He started 15 years ago with those paralyzed rats, putting tiny electrical implants into their spines to stimulate nerve fibers below the site of their spinal cord injuries. When the implant’s electrodes were powered up, Courtine’s team could train the rats, using a harness to support the animals while encouraging them to walk forward. “After two months of training, a rat that was completely paralyzed walked to the delicious piece of Swiss chocolate that we put at end of track,” Courtine remembers.

A rat held upright in a harness walks forward on its back legs.
Photo: EPFL
Gregoire Courtine's early studies enabled rats with paralyzed hind limbs to walk.

This miraculous feat was possible because the rat’s nervous system adapted to its injury with the help of the stimulation and training; the few nerve fibers around the spinal cord injury that had been spared from damage regrew and reorganized to bring commands from the rat’s brain to its legs. 

But Courtine had no intention of stopping with rats, and worked to optimize the technology in monkeys. Now his research has reached an important milestone, as the first human clinical trial of the spinal implant system has just begun at the Lausanne University Hospital. Courtine described the new trial and the research that led to it at a talk at SXSW Interactive, the massive tech festival underway in Austin, Texas, and after the talk he sat down with IEEE Spectrum to get into the technical details. 

A man shown from behind rests in a harness holding him up. A scientist stands next to him.
Photo: Hillary Sanctuary/EPFL
Gregoire Courtine [left] speaks with one of the first paralyzed patients to test out the spinal implant system.

In the clinical trial, patients use a scaled-up version of the harness the rats used, a rigging which supports them while they try to walk forward. “It holds them like a parent would hold a young child to take his first steps,” Courtine says.

The first person to get the spinal implant is a man who suffered a spinal cord injury five years ago that left him partially paralyzed. Before the implant, he was able to walk with support, but the stimulating implant drastically improved his gait. “When we turn on the stimulation, the movement is much more coordinated,” Courtine says. The trial will involve eight patients, with later patients having more severe injuries and higher levels of paralysis. 

The trial is a proof-of-concept study to show the safety of the spinal implant and the basic efficacy of the stimulation. For each patient, a surgeon places the implant on the surface of the lumbar spinal cord in the lower back. The implant is a commercial device from Medtronic that’s already approved for use in spinal stimulation therapies for chronic pain. In the operating room, the researchers make sure that the implant is properly positioned so that its 21 electrodes can stimulate nerve fibers that both extend and flex the leg muscles. The stimulator is attached to an electric pulse generator that’s nestled inside the patient’s torso.  

Once each patient has recovered from surgery, the researchers test out stimulation patterns. They turn on various electrodes at different current levels and map the effects on the patient’s leg muscles, enabling them to personalize the pattern and produce the best walking movements. But Courtine says the stimulating implant, which they chose because it already has regulatory approval, is far from ideal. “We are very frustrated by the number of electrodes,” he says.

A transparent flexible ribbon with tiny gold lines on it is held between two hands.
Photo: EPFL
The flexible "e-dura" developed by EPFL professor Stephanie Lacour is meant to interface safely with delicate neural tissue.

For the stimulator, Courtine would also vastly prefer to use the soft, flexible electrodes developed by his colleague Stéphanie Lacour, another professor of neuroprosthetics at EPFL. She has developed stretchy implantable electrodes modeled after the dura matter, the membrane that covers the spinal cord and brain. But Lacour, also at SXSW, says her “e-dura” implants won’t be ready for clinical trials for some years. 

The current human study will look for both immediate improvement to the patients’ walking ability as well as gradual improvement over five months. Courtine says he expects patients’ big strides to come from the nervous system’s remarkable neuroplasticity: “In rats and non-human primates, we see reorganization not just at site of injury, but throughout the nervous system,” he says. “The brain finds new ways to communicate with the spinal cord below the injury.”

Gif shows a monkey walking on a treadmill dragging one rear leg. Text reads "BSI off."
With the brain-spinal interface system turned off, a partially paralyzed monkey drags its rear foot.

There’s another big technological step required on the path to truly reversing paralysis. In a breakthrough study on paralyzed monkeys published last November, Courtine and his collaborators showed that inserting both a brain implant and a spinal implant provided much more natural walking movement. In that study, the brain implant in the motor cortex recorded the monkey’s intentions to move, and sent those decoded commands to the spinal stimulator.

A monkey walks normally on a treadmill. Text says "BSI on."
With the brain-spinal interface system turned on, the monkey walks almost normally.

Courtine doesn’t yet have permission to try this whole brain-spine interface in human patients, but he’s working with the international BrainGate corsortium that’s developing a clinic-ready brain implant. Using paralyzed patients’ natural brain signals is bound to bring big improvements in the control of their bodies. “Timing is so important, and the brain has perfect timing,” Courtine says. “Machines will never be able reproduce that exactly.”

Can Ultraprecise Time Measurements Warp Space?

Einstein and Heisenberg suggest that exact time measurements can produce tiny gravitational fields
Illustration: Juan Carlos Palomino/University of Vienna
Measuring time at one point can alter the flow of time in the surrounding space

File this under “fun to think about”: Researchers at the University of Vienna have shown how ultraprecise measurements of time can bend the surrounding space and make time in the region run slower.

The basic theory is surprisingly simple, a combination of Einstein’s mass-energy equivalence and Heisenberg’s uncertainty principle. In short, increasing the precision in the time measurement increases the uncertainty in the energy at that point. Since energy and mass are interchangeable, this is the same as creating a virtual mass. As the uncertainty in the time measurement falls, the “mass” increases. And as the mass increases, so does its gravity. The result is a regional gravitational time dilation—the effect that causes clocks on Earth to run slower than clocks on Global Positioning System satellites, for example.

To be sure, the effect cannot be produced or detected with today’s tools. In their paper in the Proceedings of the National Academy of Sciences , the Vienna researchers—Esteban Castro Ruiz, Flaminia Giacomini, and Časlav Brukner of the university’s Vienna Center for Quantum Science and Technology—measure the time dilation effect in “decoherence time”: The greater the warping of space-time, the smaller the decoherence time. Today’s best atomic clocks are accurate to within about 3 x 10-18 (a dimensionless number, 3 parts in 10 quintillion). Measurements to this level of accuracy produce a “mass” equal to about one ten-millionth of proton. At a distance of one or two nuclear diameters, the decoherence time is on the order of the lifetime of the universe. It’s nothing to worry about in the real world.

Clock accuracy is increasing rapidly, however, and the pace of improvement is accelerating as optical clocks and optical-lattice clocks come online. The precision of optical-lattice clocks, in particular, has increased by about four orders of magnitude in the past decade. If clock accuracy climbs to about 10-27 (less than 15 years at the current rate of improvement), the mass uncertainty grows to about 7 x 1011 electronvolts (some 350 times the mass of the proton) producing a decoherence time on the order of two minutes at a distance of 10-10 meters. At this scale, the effect might become detectable.

"Our findings suggest that we need to re-examine our ideas about the nature of time when both quantum mechanics and general relativity are taken into account", said Castro in the university’s statement on the work.

Deep Learning First: Drive.ai’s Path to Autonomous Driving

These cars use deep learning to turn past experience into better decisions
Photo provided by Drive.ai

Last month, IEEE Spectrum went out to California to take a ride in one of Drive.ai's autonomous cars, and to find out how they're using deep learning to master autonomous driving.

It's only been about a year since Drive.ai went public, but already, the company has a fleet of four vehicles navigating around the San Francisco Bay Area (mostly) autonomously—even in situations that are notoriously difficult for self-driving cars, like at night, or when it's raining.

Drive.ai structured its approach to autonomous driving entirely around deep learning from the very beginning. "This is in contrast to a traditional robotics approach,” says Sameep Tandon, one of Drive.ai’s founders. “A lot of companies are just using deep learning for this component or that component, while we view it more holistically."

Often, deep learning is used in perception, since there's so much variability inherent in how robots see the world. Many companies use deep learning for recognizing pedestrians in a camera image (to take one example), because deep learning excels at identifying one particular kind of thing (like a person) from within an arbitrary scene. Essentially, a deep learning system is able to learn to recognize patterns, and then extend that capability to patterns that it hasn't actually seen before: you don't have to train it on every single pedestrian that could possibly exist for it to be able to identify them.

While a pedestrian in a camera image is a perceptual pattern, there are also patterns in decision making and motion planning that deep learning can be applied to, and Drive.ai is leveraging deep learning here as well. For example, the correct behavior at a four way stop, or when turning right on a red light, is the kind of variable, situation-dependent decision that deep learning algorithms excel at.

Deep learning systems thrive on data. The more data an algorithm sees, the better it'll be able to recognize, and generalize about, the patterns that it needs to understand to drive safely. Data are not all created equal, though, which is why an immense amount of effort goes into collecting high quality data and then annotating it so that it's useful for training deep learning algorithms.

What differentiates Drive.ai is that it’s able to use deep learning and automation for annotating data, helping to automate the data interpretation process from the start. Drive.ai has a small team of human annotators, most of whom are kept busy training brand new scenarios, or validating the annotation that the system does on its own.

"What we want to be able to do is to train deep learning systems to help us with the perception and the decision making but also incorporate some rules and some human knowledge to make sure that it’s safe,” says Tandon.

Read More: How Drive.ai Is Mastering Autonomous Driving with Deep Learning

Intel Buys Mobileye for $15 billion

The deal rounds out two years of frenetic robocar acquisitions
Image: Mobileye

Intel is buying Mobileye, the Israeli robocar firm, for $15.3 billion. It’s one of the largest robocar acquisitions in a two-year buying frenzy that has swept both the auto industry and the tech companies that want to eat its lunch.

Mobileye  made its name selling machine vision systems for driver-assistance features, such as lane keeping and emergency stopping. Unlike many companies, notably Waymo, it has so far eschewed expensive lidar, choosing instead to depend on a single (“mono”) camera.    Mobileye has done work for most of the major car makers in the world; the most prominent—but by no means the largest—such relationship was with Tesla Motors, which  ended with some acrimony  last year.

Intel has thus bought itself not only a full suite of robocar technology but also wide-ranging contacts in the auto industry. Its newly established self-driving unit also incorporates a 15 percent stake, which Intel acquired last month , in Here, a mapping company that BMW, Daimler and Volkswagen bought from Nokia in 2015 for $2.6 billion.

Intel’s self-driving unit will be run by Mobileye management, in Jerusalem.

The deal validates Mobileye’s strategy of resisting the found-and-flip routine that most Israeli tech startups have followed (though that trend may now be changing). It held out for top dollar and it got it: Not only is $15 billion the third-largest market valuation of any publicly traded Israeli company, it is high for an auto-parts supplier and not unrespectable for an OEM.

Look at Mazda’s market capitalization, which today stands at $8.7 billion. Or at the $8 billion that Samsung paid in November for Harmon, the car-audio tech company. The only deal that comes to mind that was comparable in size was NXP’s $12 billion purchase of Freescale, an auto chip maker, back in 2015.

By comparison, Uber spent just $680 million last August to acquire Otto, a self-driving truck startup. At the time, Uber said that it valued the expertise of Otto’s staff, above all the veteran robocar expert  Anthony Levandowski, a key founder of what is now Waymo.

Uber thus had to pay just 4.4 percent as much as Intel just did for what might be called a roughly comparable suite of robocar knowledge. Uber got quite a bargain—you might even say it was a steal.