Google Maps’ new features at I/O 2021 are all about the AI

i 1 how google maps is incorporating ai

Each time you ask Google Maps to supply driving instructions, it considers many choices and selects one as the optimum route. Naturally, getting you to your vacation spot in an environment friendly method is a major aim. However if you set out on a visit, effectivity isn’t the single most necessary issue. Above all, you’d prefer to get there safely.

That’s the premise behind a new characteristic that Google unveiled at present throughout this yr’s on-line model of its I/O developer conference. Google Maps will now establish highway segments the place drivers are inclined to slam on their brakes. It is going to attempt to route you round such areas even when they’re theoretically a part of the most blatant route.

Determining the place the hazard zones are so you may keep away from them is “considered one of the most advanced issues I’ve been fortunate sufficient to deal with in my time at Google,” says director of product Russell Dicker, who’s labored on Maps on and off for seven years. The corporate solved it by making use of AI to knowledge, because it’s been doing with a bevy of different latest and upcoming tweaks to the world’s hottest mapping app.

It’s fairly apparent why laborious braking could be an indication of harmful stretch of highway: It’s proof that drivers are reacting to one thing sudden. And if everybody concerned doesn’t react shortly sufficient, the outcome could be an accident. Certainly, Dicker says that the inspiration for the new Google Maps characteristic got here from an incident a few years in the past when a Google Maps product supervisor rear-ended his father’s automobile at “this intersection with a type of super-short yellow lights.” Everybody was okay, however the mishap led the Googler to delve into the matter of hard-braking incidents— the topic of appreciable analysis by organizations resembling the Virginia Tech Transportation Institute.


We predict that we’re going to have the capacity to doubtlessly remove round 100 million hard-braking occasions.”

Russell Dicker, Google

The extra the Maps staff seemed into the points that may result in laborious braking—which vary from highway geometry to daylight hitting drivers in the eyes—the extra snug it felt factoring them into its driving instructions. “We’ve seen that there generally is a sudden improve in hard-braking occasions alongside a section when it’s raining further laborious,” says Dicker. “And so this was the subsequent ‘Aha’ second for us, as a result of we realized that understanding environmental elements and serving to individuals navigate them efficiently was what Google Maps has carried out for years.”

So how does one establish roadways that are liable to hard-braking incidents? Google had an apparent alternative to gather related knowledge: The Google Maps app runs on smartphones geared up with accelerometers, permitting it to detect movement or the abrupt lack thereof. However telephones aren’t bolted to autos; they’re topic to unbiased motion of their very own inside the cabin. That meant that uncooked accelerometer knowledge was of restricted worth.

Google found a workaround in the reality {that a} first rate chunk of Google Maps navigation includes Android Auto—the characteristic, constructed into many latest autos, that permits you to challenge apps out of your cellphone onto a dashboard touchscreen. A cellphone that’s powering an Android Auto session is at least tethered to the automobile it’s in, and Google discovered that it offered extra strong proof of laborious braking. Through the use of this knowledge to construct a machine-learning mannequin after which extrapolating for the bigger knowledge set of movement knowledge from Google Maps apps on telephones inside different autos, the firm was in a position to construct a database of laborious braking-prone areas that meant one thing.

“There’s nobody place the place half the time you drive there, you’re going to have a hard-braking occasion,” says Dicker. “The world isn’t constructed like that. And so it’s [about] looking for the actually delicate issues. That’s why the great amount of information is tremendous useful, as a result of we will actually attempt to detect these patterns.”

The information ensuing from this AI-infused course of is historic in nature slightly than in actual time. That signifies that it’s not going to assist individuals keep away from accidents in progress—one thing Google Maps already does by stay visitors knowledge. However it may possibly steer drivers away from drawback spots whose dangers could haven’t even been readily obvious till now. And on condition that Google Maps has greater than a billion customers, the cumulative impact might be profound.

“We predict that we’re going to have the capacity to doubtlessly remove round 100 million hard-braking occasions on routes pushed with Google Maps yearly,” says Dicker.

For these on foot

Together with motorists, one other necessary Google Maps constituency is pedestrians and others touring primarily by the use of sidewalks. They too wish to get from level A to level B—however what issues alongside the means is strikingly completely different from the wants of these in autos. As an example, if you’re strolling in a busy space, you care about crosswalks. And should you’re utilizing a wheelchair or mobility scooter, or pushing a stroller, that you must know the place to search out cuts in the curb.


Amassing and conveying such particulars was a key motivation behind the detailed new maps that Google started rolling out final yr, beginning with London, New York, San Francisco, and Tokyo. Now it’s saying plans to succeed in a complete of fifty areas—together with Berlin, São Paulo, Seattle, and Singapore—by the finish of this yr.

Oren Naim—a 14-year Google veteran and, like Dicker, a Google Maps director of product—acknowledges that increasing protection to even 50 cities could sound like a smallish whoop. “The factor is, technically talking, it’s extremely laborious to make these sorts of high-fidelity maps,” he stresses. The methods that Google used to get to 50 cities can scale means up. And as with the hard-braking avoidance characteristic, they have been potential solely due to AI.

To suss out further element for its maps, Google educated AI to identify highway components in aerial, satellite tv for pc, and Avenue View imagery. [Image: courtesy of Google]

To search out the pedestrian-friendly components it needed so as to add to its maps, Google turned to photographic imagery, together with satellite tv for pc and aerial images in addition to its personal Street View. At first, human beings eyeballed the pictures for the objects in query and recreated them inside Google Maps. However “there’s no means that you are able to do that for the complete world,” says Naim. “And so what we would have liked to do is to discover a approach to train a machine basically to attract these maps for us.”

In recent times, tech firms resembling Google have made huge strides in utilizing machine studying to coach computer systems to establish components in pictures at scale. However the objects the Maps staff wanted to pinpoint offered some particular challenges. A cat, for example, is at all times a cat; one thing like a crosswalk, nonetheless, can differ wildly relying on the place you are.

“For us, crosswalks have this white, black, white, black form of sample,” says Naim. However “should you go to London, crosswalks are really these parallel dots. Round the world, you’ll see these sorts of highway particulars change, not even simply at a rustic degree however at a state and generally even a metropolis degree.” Then once more, as Google educated its fashions, it additionally found consistency in some shocking locations—Atlanta and Ho Chi Minh Metropolis, Vietnam, for example, had sidewalks and visitors lights that resembled one another. Who knew?

London’s dotted crosswalks are a hanging departure from their striped U.S. counterparts. [Photo: courtesy of Google]

One other side of the job that sophisticated issues: The imagery Google was working with wasn’t designed for the functions the firm had in thoughts. For instance, Google was attempting to make use of aerial images to identify paths in parks. However parks have timber, and timber are tall, so the paths might be robust to identify. The answer entailed “fusing collectively completely different sources of images to create a mannequin that reveals you the identical characteristic from a number of angles,” explains Naim.

In the finish, people nonetheless had to assist the AI algorithms accurately deduce what they have been seeing. However the progress they made can pay dividends as Google strives to carry the new element to a whole bunch—after which 1000’s—of cities. The necessary factor right here is that if you attempt to repair it, you repair it not simply by cleansing this occasion, however you be taught one thing new about the world and you employ that data for the subsequent metropolis,” says Naim. “And they also turn out to be higher and higher over time.”

The hard-braking detection and newly detailed maps are amongst Google Maps’ most bold AI-powered additions, however they’re solely two of a number of new Maps features being unveiled at I/O. (One other one goals to be smarter about the companies that get known as out in a map—breakfast joints in the morning, for example, and dinner spots at evening.) All advised, Google believes that it’s on track to deploy over 100 AI-enabled enhancements to Maps in 2021. We’re more and more attempting to construct new sorts of information about what’s taking place in the world into all of our experiences,” says Dicker. Which signifies that it could be extra sensible to maintain monitor of modifications to Maps that don’t contain AI.