Video game tech at play in self-driv­ing cars

Northwest Arkansas Democrat-Gazette - - BUSINESS & FARM - BRIAN FUNG

Self-driv­ing cars are get­ting smarter all the time. Ad­vances in soft­ware and ar­ti­fi­cial in­tel­li­gence mean the machines are now able to dis­tin­guish be­tween cars and cy­clists, or be­tween pedes­tri­ans and a pet. Many can now “see,” pick­ing out ob­jects and ob­sta­cles ap­proach­ing ahead. All that tech could even­tu­ally save lives, help­ing to pre­vent the 95 per­cent of car ac­ci­dents that safety reg­u­la­tors es­ti­mate are caused by hu­man er­ror each year.

But none of this would be pos­si­ble with­out a piece of hard­ware many take for granted. It’s a tech­nol­ogy that traces back to the ear­li­est days of modern per­sonal com­put­ing, one that peo­ple tend to as­so­ciate more with World of War­craft than new­fan­gled wid­gets on wheels.

We’re talk­ing about the graph­ics pro­ces­sor.

In main­stream per­sonal com­put­ers, the graph­ics pro­ces­sor — of­ten found on a graph­ics card — is what al­lows com­put­ers to draw all those pix­els and poly­gons that make up today’s photo-re­al­is­tic video games. But as these pro­ces­sors have grown ever more pow­er­ful, engi­neers have dis­cov­ered their util­ity in all sorts of nongam­ing ap­pli­ca­tions. Graph­ics pro­cess­ing units — or GPUs — have tran­scended their ori­gins to be­come en­tire com­put­ers in their own right.

“[The GPU] is now pow­er­ing ev­ery­thing from games to the vis­ual ef­fects you see in Hol­ly­wood films,” said Danny Shapiro, the se­nior di­rec­tor of au­to­mo­tive at Nvidia, which ac­counts for roughly 75 per­cent of the $7.8 bil­lion mar­ket for graph­ics pro­cess­ing units. GPUs, said Shapiro, are cen­tral to “pro­fes­sional graph­ics, for au­tomak­ers that are de­sign­ing cars, to doc­tors and re­searchers that are search­ing for cures for can­cer and us­ing med­i­cal imag­ing tech­niques.”

It’s a sign of how big the GPU busi­ness has grown that some 200 other com­pa­nies work with Nvidia’s au­to­mo­tive unit alone. GPUs are even part of the brains be­hind ar­ti­fi­cial in­tel­li­gence, ap­pear­ing in tech­nolo­gies such as the Ama­zon Echo, which con­verts nat­u­ral hu­man speech into data that machines can un­der­stand.

“The com­bi­na­tion of GPUs and a CPU are now avail­able that can ac­cel­er­ate an­a­lyt­ics, deep learn­ing, high-per­for­mance com­put­ing, and sci­en­tific sim­u­la­tions,” Chris Niven, re­search di­rec­tor for oil and gas is­sues at the re­search firm IDC, told ZDNet last month.

Tra­di­tion­ally, the brain in most per­sonal com­put­ers has been the CPU, or the cen­tral pro­cess­ing unit. These chips are made by com­pa­nies such as In­tel. Ap­ple also has been mak­ing its own, pro­pri­etary chips for the iPad and iPhone. The dis­tin­guish­ing fea­ture of this tech­nol­ogy is that it’s de­signed to run cal­cu­la­tions se­ri­ally, one af­ter an­other, very quickly. The rise of dual- and quad-core CPUs have ex­panded their ca­pa­bil­i­ties, al­low­ing for more com­pu­ta­tions to oc­cur si­mul­ta­ne­ously.

These chips are still ideal for machines that only need to run a few pro­cesses at the same time. But when it comes to tech­nol­ogy such as self-driv­ing cars, where the com­put­ers are con­stantly re­ceiv­ing and di­gest­ing in­for­ma­tion, mul­ti­task­ing be­comes that much more im­por­tant. And that’s where GPUs ex­cel.

Com­puter re­searchers be­gan to dis­cover the po­ten­tial be­hind GPUs as far back as the late 1990s, when the mar­ket was awash with dozens of com­pet­ing chip­mak­ers. Their prod­ucts found their way into desk­top com­put­ers and gam­ing con­soles like the Sega Dream­cast and Xbox, en­abling con­sumers to ex­pe­ri­ence ground­break­ing ti­tles like Half-Life, Quake and Halo. By si­mul­ta­ne­ously and ef­fi­ciently con­trol­ling the gen­er­a­tion of shapes on a screen, GPUs helped bring first vector graph­ics, and then in­di­vid­ual pix­els, to life.

One pa­per in 2002 found that com­pared with CPUs, “the graph­ics hard­ware al­lows us to estab­lish a high-speed cus­tom data pro­cess­ing pipe­line. Once the pipe­line is set up, data can be streamed through with dev­as­tat­ing ef­fi­ciency.”

This is why self-driv­ing cars find GPUs so use­ful. Through the use of cameras, laser and radar sen­sors, cars look at their sur­round­ings by tak­ing many mea­sure­ments per sec­ond.

“It’s 30 pic­tures ev­ery sec­ond,” Shapiro said. “Each pic­ture, a sin­gle frame, is made up of pix­els. Each of these pix­els or dots is a nu­mer­i­cal value that says, ‘What is the color of the light there?’ It’s just a bunch of num­bers.”

GPUs like the ones found in self-driv­ing cars are de­signed to crunch those num­bers and fig­ure out that some of those pix­els rep­re­sent an ob­sta­cle, whereas other pix­els are lane mark­ings and still oth­ers are traf­fic lights. While GPUs weren’t orig­i­nally in­vented for those pur­poses, car engi­neers be­gan tak­ing ad­van­tage of the tech­nol­ogy’s par­al­lel com­put­ing pow­ers about six or seven years ago, ac­cord­ing to Ped­die.

“The orig­i­nal use of GPUs in an au­to­mo­bile was for the in­stru­ment panel in the en­ter­tain­ment sys­tem,” he said. “It’s only been re­cently that peo­ple have been say­ing, ‘Hey, we can do this, or that!’”

Within au­to­mo­biles, many stand-alone pro­ces­sors that once han­dled a sin­gle func­tion — such as the anti-lock brakes or the power win­dows — will all some­day be routed through a sin­gle pro­ces­sor, the GPU, said Shapiro. And cars in­creas­ingly will work like Tesla’s au­to­mo­biles, where driv­ers might cus­tom­ize ve­hi­cles by pick­ing and choos­ing dif­fer­ent soft­ware pack­ages to suit their driv­ing style.

“You can al­most have in-app pur­chases to add new fea­tures that weren’t there when you bought it,” he said.

For gamers who’ve grown ac­cus­tomed to buy­ing ex­pan­sion packs to their soft­ware — also known as down­load­able con­tent — this idea might sound very fa­mil­iar.

Bloomberg News/MAR­LENE AWAAD

A Google self-driv­ing car drew at­ten­tion last year at a tech­nol­ogy con­fer­ence in Paris. The de­vel­op­ment of graph­ics pro­ces­sors has led to the more pow­er­ful com­put­ers needed to au­to­mate cars.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.