Au­to­mo­bile PLUS

THREE TECH TRENDS THAT ARE HELP­ING DRIVER­LESS CARS THROUGH THEIR EARLY YEARS

Automobile - - Contents - By Doug New­comb

What are au­tomak­ers work­ing on now to speed up the fu­ture of the car? We find out. We also let you know about some fun gad­gets for cars of all ages.

Re­cently teach­ing my teenage son to drive, I couldn’t ex­plain in­stinc­tive be­hav­ior I take for granted af­ter 40-plus years

be­hind the wheel.

While things like fol­low­ing too closely and brak­ing too late were easy to point out, an­tic­i­pat­ing the ac­tions of other driv­ers and deal­ing with com­plex in­ter­sec­tions only come with ex­pe­ri­ence.

Au­ton­o­mous ve­hi­cles (AVs) are like new­bie driv­ers, ex­cept with bet­ter-de­vel­oped brains and bil­lions of dol­lars in tech to help shorten the learn­ing curve. But even with all their sen­sors and soft­ware, AVs still have flaws to over­come be­fore they drive with com­plete con­fi­dence and com­pe­tence.

In the race to get robo­cars on the road, sev­eral un­der­the-radar tech trends are co­a­lesc­ing to help make true AVs a re­al­ity—and maybe make my son’s gen­er­a­tion one of the last to learn to drive it­self.

Teach­ing AVs the Rules of the Road

My new driver had to study the Ore­gon Driver Man­ual to learn the dif­fer­ence be­tween, say, a stop and yield sign. AVs sim­i­larly learn the rules of the road, but through ar­ti­fi­cial in­tel­li­gence (AI). An­other re­quire­ment for my son is to log 50 hours with an adult in the car. AVs also ac­quire real-world ex­pe­ri­ence by putting in hours on the road, but only in cer­tain lo­ca­tions and con­di­tions be­cause of le­gal re­stric­tions and weather.

AVs learn to in­ter­pret signs and other road­way info via a type of AI known as ma­chine learn­ing, which re­quires driv­ing a route and hu­mans ver­i­fy­ing the data. Traf­fic­data com­pany In­rix has a way for AVs to more quickly learn the rules even in places they’ve never driven. In­rix’s AV Road Rules plat­form lets cities dig­i­tize their traf­fic in­fra­struc­ture and rules. This not only cre­ates a short­cut for AVs to mem­o­rize traf­fic rules but also al­lows them to op­er­ate from ac­cu­rate data.

“For 100 years, signs and lane mark­ings have been the lan­guage of com­mu­ni­cat­ing traf­fic rules to driv­ers, and it’s worked pretty well,” says Avery Ash, head of au­ton­o­mous mo­bil­ity for In­rix. “But we’ve all been in sit­u­a­tions where the sig­nage is con­fus­ing or ob­scured or lane strip­ing has been worn off, but we fig­ure it out.”

Al­though ma­chine learn­ing can help AVs fig­ure out such sit­u­a­tions, Ash adds that “it’s a te­dious, lengthy, and ex­pen­sive process, and the re­sults are not ac­cu­rate enough

for the sort of safety-crit­i­cal op­er­a­tion re­quired by AVs. AV Road Rules is an ad­di­tional data layer that com­ple­ments ma­chine learn­ing and HD maps.”

Once AVs know the rules, they need to ap­ply them on roads. But just as I don’t have 50 ex­tra hours to spend driv­ing with my son, AV op­er­a­tors have lim­ited time and re­sources when log­ging miles.

A 2016 RAND study es­ti­mated that AVs “would have to be driven hun­dreds of mil­lions of miles and some­times hun­dreds of bil­lions of miles to demon­strate their safety in terms of fa­tal­i­ties and in­juries.” But de­vel­op­ers have found ways to speed up the process through sim­u­la­tion soft­ware, and most are us­ing sim­u­la­tion to ac­cel­er­ate AV de­ploy­ment.

While Waymo leads the pack in terms of real-world miles trav­eled test­ing its AVs—more than 9 mil­lion since par­ent com­pany Google first kicked off the project al­most a decade ago—it’s able to achieve nearly the same num­ber of miles ev­ery day us­ing sim­u­la­tion soft­ware. “Hav­ing a proper sim­u­la­tion strat­egy is the only way that au­ton­o­mous ve­hi­cles can train the sen­sors and de­ci­sion-mak­ing func­tions for road test­ing and have the con­fi­dence that their tech­nol­ogy is safe and ready,” says Danny Atsmon, CEO of Cog­nata, which is work­ing with Audi and oth­ers.

Sim­u­la­tion also al­lows AV de­vel­op­ers to test for “edge cases” such as pedes­tri­ans and cy­clists sud­denly cross­ing in front of the car or when the sun shines di­rectly into an AV’s front-fac­ing cam­era at sun­set, tem­po­rar­ily blind­ing it.

“With sim­u­la­tion, we’re able to recre­ate blind­ing sun 24 hours a day,” says Danny Shapiro, se­nior di­rec­tor of au­to­mo­tive for chip­maker Nvidia. “And we can do it on ev­ery road and com­bine that with a rain­storm or any kind of weather. We can also sim­u­late a car run­ning a red light and eval­u­ate if the AV is tak­ing the cor­rect ac­tion and de­tect­ing every­thing it should.”

Re­mote Pos­si­bil­i­ties

Even with all their com­put­ing power and AI, sit­u­a­tions re­main that re­quire AVs to call on hu­man driv­ers for help. That’s why most AVs test­ing on pub­lic roads need a hu­man be­hind the wheel to take over when a self-driv­ing com­puter gets con­fused or can’t con­tinue for some rea­son, such as when a road is closed or there’s a tem­po­rary con­struc­tion zone.

But as AVs shed com­po­nents such as steer­ing wheels and ped­als and hu­mans be­come cargo, they’ll need to be re­motely con­trolled; Cal­i­for­nia’s reg­u­la­tions for test­ing AVs on pub­lic roads that took ef­fect in April re­quire so. En­ter tele­op­er­a­tion, the in­dus­try term for re­motely op­er­ated AVs.

“Think of tele­op­er­a­tion as an air traf­fic con­troller for au­ton­o­mous ve­hi­cles,” says Jada Smith, VP of ad­vanced en­gi­neer­ing at Ap­tiv (formerly Del­phi), which runs a rob­o­taxi pro­gram in Las Vegas in part­ner­ship with Lyft. But un­like air traf­fic con­trol, AV tele­op­er­a­tors will be able to not only mon­i­tor but also op­er­ate self-driv­ing cars when an au­ton­o­mous ve­hi­cle en­coun­ters a sit­u­a­tion it doesn’t know how to han­dle.

Most ma­jor AV play­ers are ei­ther pre­par­ing for tele­op­er­a­tion of robo-taxis or test­ing it al­ready. GM’s sta­ble of Chevy Bolts be­ing retro­fit­ted by its Cruise Au­to­ma­tion sub­sidiary to op­er­ate with­out a steer­ing wheel or ped­als have an “expert mode,” which re­lies on tele­op­er­a­tor as­sis­tance. Toy­ota has a pa­tent for “re­mote op­er­a­tion of au­ton­o­mous ve­hi­cle in un­ex­pected en­vi­ron­ment,” while self-driv­ing startup Zoox has one for a “tele­op­er­a­tion sys­tem and method for trajectory mod­i­fi­ca­tion of au­ton­o­mous ve­hi­cles.”

NASA re­motely con­trolled a se­ries of Mar­tian rovers in the late 1990s and early 2000s. That’s why Nissan, mean­while, has re­cruited for­mer NASA sci­en­tists to ap­ply a ver­sion of the space agency’s tele­op­er­a­tions ex­per­tise to the au­tomaker’s au­ton­o­mous ve­hi­cles and is test­ing re­mote con­trol of a fleet of self-driv­ing Leaf EVs at NASA’s Ames cam­pus in Sil­i­con Val­ley.

In the first demon­stra­tion of the tech­nol­ogy on pub­lic roads (and on Earth), at the Con­sumer Elec­tron­ics Show last Jan­uary, Phan­tom Auto had a hu­man op­er­a­tor 500 miles away in Cal­i­for­nia con­trol a car driv­ing on the Las Vegas Strip. “AV tech­nol­ogy may be about 97 per­cent of the way there, but that last 3 per­cent may take decades to solve,” says El­liot Katz, co-founder and chief strat­egy of­fi­cer for Phan­tom Auto. “Tele­op­er­a­tion … serves as an es­sen­tial tech­no­log­i­cal bridge which en­ables AVs to be safely de­ployed now.”

Watch Me Now

Along with re­mote mon­i­tor­ing of self-driv­ing cars, hu­man driv­ers will also in­creas­ingly be un­der scru­tiny, es­pe­cially in the in­terim be­tween SAE Lev­els 3 and 4 of au­ton­omy when hu­mans will need to be ready to take con­trol at a mo­ment’s no­tice. Al­though cam­eras are al­ready used in some cars to mon­i­tor driv­ers, such as with Cadil­lac Su­per Cruise, a new gen­er­a­tion of cam­eras will move be­yond sim­ply de­tect­ing whether the driver’s head is turned away from the road to in­clude facial recog­ni­tion and even be able to read emo­tions of pas­sen­gers in fully au­ton­o­mous cars.

Subaru in­tro­duced a fea­ture on its all-new Forester, called DriverFo­cus, that uses facial recog­ni­tion soft­ware to look for signs of driver dis­trac­tion and fa­tigue. Part of the Subaru Eye­Sight suite of driver as­sists, DriverFo­cus can de­tect when some­one is doz­ing off or look­ing away from the road for too long, and it will au­to­mat­i­cally stop the ve­hi­cle.

An­other com­pany, eye­Sight (no re­la­tion to the Subaru op­tion), has a cam­era that not only de­tects dis­trac­tion and drowsi­ness but also in­cludes what the com­pany calls “con­tex­tual con­trol” based on the di­rec­tion of a driver’s gaze to high­light con­tent in cock­pit dis­plays. A creepier pro­fil­ing fea­ture al­lows it to de­tect the age and gen­der of the driver and use the info for “con­nected car an­a­lyt­ics.”

Ren­ovo Auto is fo­cused on cre­at­ing a uni­ver­sal op­er­at­ing sys­tem for AVs called aWare. It in­cor­po­rates sen­sor and soft­ware tech­nolo­gies and in­volves us­ing cam­eras in­side and out­side a ve­hi­cle to cap­ture emo­tions of pas­sen­gers as well as pedes­tri­ans.

“Dur­ing de­vel­op­ment, you need to mon­i­tor driv­ers to make sure they are en­gaged,” Ren­ovo CEO Chris Heiser says. “And later you’ll need to in­ter­act with the pas­sen­gers to help build trust and pro­vide them all the ser­vices that a hu­man driver does today.”

To pro­vide this interaction, Ren­ovo is work­ing with AI star­tups Af­fec­tiva and Speak With Me to in­te­grate their tech­nol­ogy that an­a­lyzes the facial ex­pres­sions and voices of pas­sen­gers into the com­pany’s fleet of AV test ve­hi­cles.

Now, if it could only make my teenage son more re­spon­sive and per­sonal when I’m in the car with him for those 50 hours. AM

Soft­waresuch as Nvidia’s Drive Con­stel­la­tion(above) cre­ates a vir­tual world to put AV sys­tems through their paces; Audi em­ploys a sim­u­la­tion plat­form from Cog­nata (leftand be­low).

Cam­eras from com­pa­nies such as Af­fec­tiva (above) and eye­Sight (be­low)de­tect the emo­tional state of pas­sen­gers based on facialex­pres­sions.

Tele­op­er­a­tion by com­pa­nies such as Phan­tom Auto (above and left) al­lows a hu­man to re­motely drive a ve­hi­cle in sit­u­a­tions when the on­board tech needs help, such as a tem­po­rary road clo­sure.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.