A driver­less car’s de­light­ful—un­less the weather out­side is fright­ful

“Snow smoke” dis­ables sen­sors ro­bot ve­hi­cles need to nav­i­gate “When it starts build­ing up, you just lose func­tion­al­ity”

Bloomberg Businessweek (Asia) - - CONTENTS -

In Jokkmokk, a ham­let just north of the Arc­tic Cir­cle in Swe­den, Volvo Cars’ self-driv­ing XC90 sport-util­ity ve­hi­cle met its match: frozen flakes that caked on radar sen­sors es­sen­tial to read­ing the road. Sud­denly the SUV was blind. “It’s re­ally dif­fi­cult, es­pe­cially when you have the snow

smoke from the car in front,” says Mar­cus Rothoff, di­rec­tor of Volvo’s au­ton­o­mous-driv­ing pro­gram. “A bit of ice you can man­age. But when it starts build­ing up, you just lose func­tion­al­ity.” So Volvo en­gi­neers started mov­ing the sen­sors. And next year, when Swedish driv­ers take their hands off the wheel of leased XC90s in the world’s first pub­lic test of fully au­ton­o­mous tech­nol­ogy, the radar will be nes­tled be­hind the wind­shield, which wipers keep clear of ice and snow.

As au­tomak­ers race to get ro­bot cars on the road, they’re en­coun­ter­ing an ob­sta­cle very fa­mil­iar to hu­mans: Old Man Win­ter. Sim­ple snow can ren­der the most ad­vanced com­put­ing power use­less. That’s why com­pa­nies in­clud­ing Volvo Cars, owned by China’s

Zhe­jiang Geely Hold­ing Group, and Ford are revving up ef­forts to pre­vent snow blind­ness.

With about 70 per­cent of Amer­i­cans liv­ing in the snow belt, learn­ing to nav­i­gate in rough weather is cru­cial for driver­less cars to gain mass ap­peal and re­al­ize their po­ten­tial to re­duce traf­fic con­ges­tion and road deaths. “If your vi­sion is ob­scured as a hu­man in strong flur­ries, then vi­sion sen­sors are go­ing to en­counter the ex­act same ob­sta­cles,” says Jeremy Carl­son, an au­ton­omy spe­cial­ist at IHS Au­to­mo­tive.

Driver­less cars “see” the world around them us­ing data from cam­eras, radar, and lidar, which bounces laser light off ob­jects to as­sess their shape and lo­ca­tion. High-speed pro­ces­sors crunch the data to pro­vide 360-de­gree de­tec­tion of lanes, traf­fic, pedes­tri­ans, signs, stop­lights, and any­thing else in the ve­hi­cle’s path. But snow can shroud cam­eras and cover the lane lines they must de­tect to keep a driver­less car on course. Lidar also is lim­ited be­cause the light pulses it emits re­flect off flakes, po­ten­tially con­fus­ing even a mod­est cur­tain of fall­ing snow with an ob­ject to avoid, caus­ing the car to hit the brakes.

Radar, which senses ob­jects by emit­ting elec­tro­mag­netic waves and has been used since 1999 in adap­tive cruise con­trol to main­tain a set dis­tance from other ve­hi­cles, is bet­ter. “If ev­ery­thing else fails, I can fol­low the pre­ced­ing traf­fic,” says Kay Step­per, vice pres­i­dent and head of the au­to­mated driv­ing unit at Ger­man parts sup­plier

Robert Bosch. “The radar is the key

el­e­ment of that be­cause of its abil­ity to work ro­bustly in in­clement weather.”

Ford, which de­clined in­ter­view re­quests, says it might have found a so­lu­tion to snow-blan­keted lane lines, ac­cord­ing to a press re­lease. The au­tomaker is scan­ning roads in ad­vance with lidar to cre­ate high-def­i­ni­tion 3D maps that are more ac­cu­rate than im­ages from global po­si­tion­ing satel­lites, which can be 10 me­ters off.

Ryan Eus­tice, an as­so­ciate pro­fes­sor of en­gi­neer­ing at the Univer­sity of Michi­gan who’s worked with Ford on the prob­lem since 2012, says they’ve also found a way to fil­ter the “noise” cre­ated by fall­ing snowflakes. The fil­tered data com­bined with in­for­ma­tion from the 3D maps al­low the car to pin­point its lo­ca­tion to within “tens of cen­time­ters,” he says. “That’s high enough ac­cu­racy that we know ex­actly what lane we’re in.”

Still, lane lines can be­come mean­ing­less in a snow­storm, as hu­man driv­ers blaze their own trails in the ruts cre­ated by ve­hi­cles in front of them. The so­lu­tion may be ar­ti­fi­cial in­tel­li­gence, says Danny Shapiro, se­nior di­rec­tor of au­to­mo­tive at pro­ces­sor maker Nvidia, which says its lat­est com­puter brain can per­form as many as 24 tril­lion “deep learn­ing op­er­a­tions” per se­cond. Deep learn­ing trains a ro­bot car, based on mil­lions of miles of driv­ing ex­pe­ri­ence loaded into its soft­ware and con­tin­u­ally up­dated. So in a snow­storm, the car will know it should fol­low the ruts rather than stay within the lane lines. “It’s very sim­i­lar to how a hu­man learns, by ex­pe­ri­ence,” Shapiro says.

Also like a hu­man, a driver­less car can get dis­ori­ented in a white­out. “There’s been a lot of hype in the me­dia and in the pub­lic mind’s eye” about the tech­nol­ogy for self­driv­ing cars “be­ing nearly solved,” Eus­tice says. “But a car that’s able to do na­tion­wide, all-weather driv­ing, un­der all con­di­tions, that’s still the holy grail.” −Keith Naughton

The bot­tom line About 70 per­cent of Amer­i­cans live in the snow belt. So car­mak­ers are try­ing to de­velop driver­less ve­hi­cles that can “see” in snow.

Newspapers in English

Newspapers from Australia

© PressReader. All rights reserved.