Look Ma, No Driver!
– driverless cars

May 6th, 2011

I don't have a driver's license. I have no plans to get one. I'm constantly (constantly) nagged about when I'll cave and just take the lessons because, allegedly, one day I'll need it. But if I hold out another decade, I just might not. Car makers intend to take out the weak point: the driver.

A staple of the sci­ence fic­tion met­ro­polis is the driver­less car: autonom­ous taxis or fam­ily cars without a steer­ing wheel are a 22nd cen­tury signum, but driver­less cars have already come a long way in design, tech­no­logy and imple­ment­a­tion. Car makers such as Volvo, Toyota and Gen­eral Motors are work­ing hard to improve traffic flow, safety and fuel effi­ciency by design­ing away the cause of most prob­lems: human drivers.

Lombard Street, San Francisco

Lom­bart Street, San Fran­cisco — one of the trick­i­est roads around

Bet­ter than you’d guess

Gen­eral Motors prom­ises I will win the no-driver’s license argu­ment within the dec­ade. Efforts for driver­less cars are turn­ing out well: from the self-parking Lexus to the Google car (200,000 kilo­met­ers, includ­ing Lom­bard Street in San Fran­cisco, with only one minor acci­dent). Driver­less cars are per­ceived as mak­ing slow mis­takes in an open park­ing lot, but con­sider the win­ners of the Vis­lab Inter­con­tin­ental Autonom­ous Chal­lenge.

When four vans rolled into the Shang­hai Expo in 2010, they’d barreled across the most remote parts of Kaza­kh­stan and Rus­sia, across vast expanses where no maps are avail­able, where the research team had no pos­sib­il­ity to plan the route before­hand. But surely, crash­ing across the empty Rus­sian waste­lands isn’t a real­istic chal­lenge? No, but these (elec­tric) cars had driven them­selves from Parma, Italy, out through East­ern Europe, to Shang­hai in three months – a jour­ney of 15,000 kilo­met­ers, with unharmed cargo. The implic­a­tion of driver­less cars is enorm­ous: autonom­ous, green, inter­con­tin­ental cargo convoys.

The road taken in the Vislab Intercontinental Autonomous Challenge

The road taken in the Vis­lab Inter­con­tin­ental Autonom­ous Challenge

Coun­ter­ar­gu­ments (countered)

There are sound reas­ons to dis­be­lieve the driver­less cars: the tech­nical chal­lenges are obvi­ous, but legal and social issues are non-trivial too:

  • algorithmic drivers aren’t cre­at­ive enough to deal with crit­ical risk scen­arios they haven’t seen before.
  • your trashy PC can’t even keep track of your email, much less nego­ti­ate a four-way inter­sec­tion – no one will bet their life against soft­ware bugs.
  • when an acci­dent hap­pens; is the pas­sen­ger, hard­ware man­u­fac­turer or soft­ware ana­lyst leg­ally responsible?
  • no-one will accept shar­ing the road with an unpre­dict­able com­puter driver, unless we replace all cars simultaneously.

Car wrapped around treeSo, driver­less cars can’t handle an unknown scen­ario? Neither can you. An adrenaline-jammed spinal cord won’t pri­or­it­ize well between get­ting rammed by an oncom­ing truck and side-swiping a bicyc­list, and the brain won’t fig­ure out a safe tra­ject­ory through. Auto­matic drivers do both, and coordin­ate responses: your car and the truck can agree on where to go. If the truck is in the wrong lane, it can announce intent to cross over com­pletely into the ditch on your side. With that inform­a­tion, your car can use the truck’s lane for slow­ing down safely. In the split second a human driver had thrown up his arms and screamed, driver­less vehicles can per­form an how-do-you-do nego­ti­ation, agree on exit paths, politely dis­cuss the weather and road con­di­tions and part as friends (pos­sibly into a tree if noth­ing bet­ter can be found).

What about buggy soft­ware? The above scen­ario assumes that no pro­gram­ming error makes the driver­less car panic and veer to avoid a semi truck that isn’t even there. Bugs crop up, and in a car at full speed, reboot­ing isn’t an option. What’s to say a bug won’t  bit-flip, turn left and chicken-race the truck? Noth­ing, apart from rig­or­ous test­ing and qual­ity con­trol. Mean­while, human drivers self-administer death by oncom­ing truck every day. Unex­pec­ted scen­arios are likely human errors any­way:

…con­ceived unpre­dict­able incid­ents […] such as an elk on the road. The inter­viewees man­aged crit­ical incid­ents by com­mon sense but this concept was also used when they excused their mis­takes.
(The Con­cep­tions of Traffic Safety among Young Male Drivers, R. Lehtimäki)

And sure, the Google car has been in an acci­dent. It was hit from behind at a red light. By a human.

But, when an acci­dent does hap­pen, who is respons­ible? A driver­less car fails to stop at a red light and runs someone over. Who does the time? The ques­tion is a sub­set of the gen­eral prob­lem of defin­ing respons­ib­il­ity around robot­ics (slides here). Peter Asaro con­cludes that “other dif­fi­cult prac­tical and meta-ethical prob­lems can­not be solved by legal the­ory alone”. Legal code has to be adjus­ted to emer­ging tech­no­lo­gies (for instance, here are six new “insane” laws we’ll need in the future). It is a dif­fi­cult topic, but ana­log­ous to legal respons­ib­il­ity for the men­tally chal­lenged or psychotic.

HAL 9000

I’m afraid I can’t let you do that, Dave.

The Low-hanging Fruit

Of course, there is the simple way: human error causes almost every acci­dent. With a driver­less car, 90% of the risk goes away, so from a game the­or­et­ical per­spect­ive there is good util­ity in choos­ing the auto­matic driver,along with full respons­ib­il­ity for its actions – you’d serve an aver­age of 90% fewer years behind bars for vehicu­lar manslaughter.

Dis­trust for machines lies within us: driv­ing into a four-way inter­sec­tion know­ing that you’re the only con­scious driver around might not inspire trust. The early adop­ter might not be appre­ci­ated: wait­ing in line all night for an iPhone 5 will upset very few, but speed­ing two tons of metal down the road past school grounds while read­ing in the back?

Charles Stross sug­gests an inter­est­ing roadmap: the adop­tion vec­tor for autonom­ous cars is rising insur­ance premi­ums. Those who insist on driv­ing manu­ally are bet­ting against the ten­fold risk, and that might become an expens­ive hobby.

Let’s get one bit of stat­ist­ics in place: cars them­selves aren’t very dan­ger­ous. Writ­ing on the “epi­demi­ology of motor vehicle col­li­sions” (lovely word­ing!), the World Health Organ­iz­a­tion (WHO) estim­ates 1.2 mega­deaths per year attrib­uted to traffic, with some fifty mil­lion injur­ies on top. Of those, ninety per­cent of all acci­dents are attrib­ut­able to driver error – emo­tion, speed­ing, inat­ten­tion and intox­ic­a­tion. That’s a cool, round mil­lion deaths a year, attrib­ut­able to human drivers. Stross, again, puts this into per­spect­ive:

If you com­pare the total cas­u­alty rates, then we have one and a half times as many people injured in road traffic acci­dents every year than were wounded or killed dur­ing the entire first world war.

Highway at night, by Michael Thurber

photo by Michael Thurber

The Romantic Car

This is social: why deprive accident-free drivers of the joy of driv­ing, just because some drivers aren’t behav­ing in traffic? This is a dis­play of Stock­holm syn­drome at best. But we love the romantic idea of the open road, sweep­ing down an empty high­way at dusk rock­ing Dire Straits’ Sul­tans of Swing, the next town unknown and the end of the road a mys­tery. Who can take out the explorer in all of us and replace it with a glor­i­fied laptop?

Three deck city, Wiley Harvey Corbett

The Three-Deck city. Illus­tra­tion via Paris Deuxieme.

Yeah, about that romantic idea of the open road… Fam­ily cars are typ­ic­ally parked for more than 90% of their lifespan. Many of them never gets to see the open, rural road, because they’re stuck in traffic. Most of them spend their car-lives in an urban environment.

Plan­ning urban areas to accom­mod­ate heavy traffic without con­ges­tion is hard. Traffic plan­ning has even fueled the devel­op­ment of math­em­at­ical fields such as game the­ory, queue the­ory and graph the­ory. Even with copi­ous amounts of plan­ning and cal­cu­la­tion, traffic is unpre­dict­able and its beha­vior is like that of vis­cous flu­ids and shockwaves.

The Car-friendly City

Since rebuild­ing the city from scratch is sel­dom an option, there have been sug­ges­tions for less dra­matic fur­nish­ings. In the 1930s, Har­vey Wiley Corbett envi­sioned a New York where rail, car and ped­es­trian traffic had been sep­ar­ated into three levels:

Since we have already put much of our rail traffic under­ground in our large cit­ies, let us elev­ate the side­walk to the exist­ing second story level and leave the street level wholly to wheel traffic.

Los Angeles is widely held as a city built for an auto­mot­ive soci­ety, but is really just filling out the space between the 30-minute car trips a fam­ily can tol­er­ate. Corbett called the res­ults of these effects old-fashioned – in 1931:

Our present side­walk sys­tem is archaic. It ori­gin­ated in a day when none could have fore­told the throngs which now stop at street inter­sec­tions to wait the passing of auto­mo­biles or dodge their way through and between them.

The con­cepts and build­ing blocks used to envi­sion our greatest cit­ies did not have heavy traffic in mind – Roman side­walks and road-crossings were ori­gin­ally inten­ded for get­ting across a muddy ditch, not to nego­ti­ate thou­sands of cars milling by.

Ped­es­trian cross­ing in Pom­peii, by Com­mon­sense Design

Even movies con­trast uto­pian futur­istic cit­ies and con­tem­por­ary cit­ies by traffic: ver­tic­ally ori­ented hous­ing, sharp sub­urban fal­loff and auto­mated, on-demand, rail­bound or elev­ated road sys­tems. If a city doesn’t suf­fer from traffic con­ges­tion, if the risk of being killed cross­ing the street is zero and if traffic swoops by on gently curving rib­bons over­head, then the city must be a met­ro­polis designed from the ground up for mod­ern auto­mot­ive traffic – a future utopia.

Future highway by Shen Fei

Illus­tra­tion by Shen Fei

The Show­down

And sure enough, we’re already partly in the future. Here’s a sweet trio of You­tube clips that nicely frame the computer-driven cars avail­able today, human drivers, and what the bleed­ing edge looks like.

1) Lexus has imple­men­ted assisted park­ing in their LS460 model. The driver des­ig­nates a box to park in, and the car auto­mat­ic­ally eases its way into the park­ing space. It’s a bit crude and looks a bit slow and care­ful, but it’s good hint of what’s com­ing. This tech is in the streets, co-existing with other cars and deemed safe and secure. That’s one big hurdle.

2) So, Ger­man power­house Audi decides to poke a bit of fun at the Lexus. In a com­mer­cial short, an Audi driver powers­lides into a nar­row park­ing spot and declares that the Audi is a “lux­ury car for people who can park themselves”.

3) Per­son­ally, I find this quip a bit on unfair side: of course an expert stunt driver can pull that of in a con­trolled envir­on­ment — it’s just not good sport to pit a human stunt driver against an autonom­ous vehicle try­ing its best.

It’s not fair to the driver. I, for one, wel­come our new driver­less overlords.


No Tweet­Backs yet. (Be the first to Tweet this post)

Leave a Reply