Self-Driving Cars and Ecology, or "Where's the Running Board?"

 NYTimes had an article on Tesla in its Sunday magazine. Its emphasis was on the problems in the self-driving software.  

Tesla claims that based on accidents per miles their cars are much safer than those driven by people.  That may well be true, but I'd love to see a test where the drivers and conditions are randomly assigned.

I think one problem is the lack of sound data--apparently each company which is trying to implement such software maintains its own data, presumably for competitive reasons. But even if the data were public, there doesn't seem to be a basis for comparison.  The testing being done uses drivers who aren't at all reflective of the overall population and is done on roads and in conditions which aren't representative of normal driving. 

As it stands the testing being done is also unfair to Tesla and the others.  What do I mean?  The current ecology of drivers, roads, and conditions has evolved over a long history. An experienced driver has expectations based on her experience, and operates on their basis. I imagine it could be modeled as a circle in a Venn diagram. Imagine 60 years from now when almost all cars are self-driving.  That ecology will have "drivers" with somewhat different expectations, cars different than todays--notably quicker to to react, and roads which will have been modified for better self-driving. In our Venn diagram, the circle for the current ecology and the circle for the self-driving ecology will not be identical; they'll overlap in some areas.

Today when we judge self-driving software we're judging it by the current ecology, not the ecology of 60 years from now.  It's like cow-catchers on locomotives or running boards on cars--both were things needed by the early rail and automotive systems, but not by the current ones (though it turns out our trucks and SUV's still need them).  It will take time for the ecology to evolve; for drivers to gain experience with the cars, for the cars and software to improve, for the roads to be modified, for the insurance industry to adapt and the laws to change. It will be evolutionary.

Tesla Self-Driving Good Enough for Me

 I struggled to find the NYTimes article I want to comment from.  Turns out it 's over a month old, on-line, but just published in the print version today. It's a ride-along with the owner of a Tesla self-driving car, something of a beta test as they try to improve their software to be safe for wide release.  The owner has a specific location which apparently gives the current software fits, a left turn with 3 lanes of on-coming traffic into a specific lane of another 3-lane road.

My reaction to the description: it's good enough for me to buy.  I've leased my last two cars, on the mistaken assumption that I might not be driving in the future (the future keeps receding, but it's coming). Anyhow, that means I know I've driven about 5500 miles in 2 years.  Granted that part of the figure is the result of reduced activity from the pandemic, but as I age I drive less.

More importantly, the proportion of my driving which occurs on routes I've used before keeps increasing.  I don't know what it is, but likely 95 percent or more.  That means if I buy a self-driving car it can quickly become familiar with my routes, and familiarity means the software can handle more and more situations.  Also, as an older driver I'm more cautious than the average bear (much to the aggravation of some of my fellow drivers), so I'm both more likely to pay attention to the car's operation and less likely to be frustrated as the car responds to new situations by slowing and/or stopping--something which I've seen reported as a problem. 

So the bottom line, there's a niche out there to sell self-driving cars; it just needs to be identified.  Once such cars get into the niche, the process of expanding their capabilities to expand the niche can proceed much faster.

Self-Driving Cars

Sunday in the print editions: 

The Post had an essay by Mr. Zipper on self-driving cars, arguing they won't save lives (at least not 94 percent of current fatalities) and won't be good for the environment.

The Times has an opinion piece by Mr. Manjoo praising the Cadillac Escalade, including its self-driving package.

Both writers seem to agree that by making driving less onerous self-driving cars will encourage more driving, more time spent in cars, and therefore hurt the environment.

I'm still looking forward to more advances, though not ready to spring for an Escalade.  But my need and interest in driving seems to be waning, particularly in the context of the pandemic. 

Autonomous Trucks

 Tyler Cowen at Marginal Revolution links to a report of Walmarts self-driving trucks, with no safety drivers. 

Self-Driving Cars Revisited

Technology Review posts a puff piece written by Baidu, a Chinese firm which seems to be progressing well in developing a self-driving car, with emphasis on steps they've taken trying to earn the trust of potential customers. 

US firms are making progress as well.  The pandemic has severely limited my driving, so it seems as if a self-driving car could  fulfill my needs soon.

My Dreams of a Self-Driving Car

 I've posted before about my hopes for a self-driving car, something which compensate for my declining physical abilities as I age. I didn't want to see the headline on this piece.

I repeat my previous suggestion of using an approach of developing a car which can memorize routes, given that many people like me do most of their driving over a limited selection of routes. Apparently from the article that's not the way companies are going--their loss.

On a personal note, I just completed my 3-year lease of a Prius with extensive safety features which saved me two or three times from likely accidents,

The Effects of the Recession and Other Stats

Three interesting stats in the news on May 29: one was how much fewer school employees the education system has today, as opposed to 2008; one was how much fewer employees the health system today as opposed to 2008; the third was how little time cars spend on roads.

Bring on Self Driving Cars

Tesla says they average 1 fatality every 3 million miles.  The government says the US averages 1 fatality every 500,000 miles.  That tells me if we gave everyone a Tesla we could save 30,000 lives a year.

That's from this NYTimes article which should have been entitled: Teslas 6 times safer than driver-driven cars.

Driverless Cars for the Disabled

An article today in the papers on someone working on driverless cars for the disabled.  It seems to me quite possible such cars will fill niche spots, long belong they become widely usable.

Will Autonomous Cars Save Parking Spaces in the City?

It's assumed the answer to my question is "yes"--some recent articles arguing for changing zoning requirements in the city to reduce the number of parking spaces required.

I'm not so sure.  As long as people commute from the suburbs to the city for their jobs, it seems to me parking is going to be a problem.  Yes, in some cases I can imagine a Reston commuter to DC getting a car at 7 for a 30 minute drive to DC, the car then returning to Reston to pick up another commuter at 8 for another 30 minute drive.  But then it's going to need to be parked until the evening.  So if the two individuals were each driving solo into the city and parking now, that would reduce the number of parking spots needed.  But that's a special pattern

Granted, you can imagine with autonomous cars and a drive sharing app, we could have much more flexible drive sharing so the number of people in the car rises from 1 to 2 or more on average.  And there likely will be realignment of jobs and homes based on the availability of autonomous cars.

My bottom line: the change requires people to change their habits, meaning it's going to take a while before the impact on parking spaces is felt. 

Driverless Cars: Setting the Bar Too High

Technology Review has a discussion of three factors impeding the adoption of driverless cars:

• safety--cars being safer than human drivers (who don't drink or text)
• useful--cars that aren't slow because too cautious, perhaps requiring regulatory changes.
• affordable.

To me it seems they're setting the bar too high.  Going back to the Innovator's Dilemma new technologies evolve by finding a niche from which they can expand gradually, making use of the learning curve to reduce costs so existing technology can be undersold and to become useful in new ways.  I think that applies here, as I've said before:

• a geezer like me isn't as safe a driver as the average person, even though we know enough not to drink or text.
• a geezer like me is already a cautious driver, so making a driverless car that abides by the speed limits is not disrupting the norm (for us).
• a geezer like me values driveability higher, highly enough to pay a premium to preserve the ability

The Future Is Near: The Impact of Autonomous Trucks

This piece at evonomics.com by Andrew Yang is distressing, and convincing. Perhaps the worst part is the impact on small businesses which serve the  trucking industry: the truckstops and motels. Yang doesn't note it, but those enterprises might well be the  lifeline for "flyover America"--the wide open spaces where farming no longer employs the people it used to.

AI, Dark Swans, Google Map Directions, and Moore's Law

Some further thoughts on the Google map episode I described yesterday.

It strikes me that AI in general will have problems with "dark swan" events.  That's true by definition--if AI is trained by using a big database of past information, it can't be trained to handle events which aren't reflected in the database.  In many cases, like autonomous cars, the algorithms can be set to do nothing.  It would be similar to the "dead man switch" found on locomotives--if there's no longer an intelligence at the controls which can react properly to events, you stop the engine.  But in other cases it may be harder to define such cases.

In cases like I encountered yesterday there may be salvation found in Moore's Law.  Presumably Google samples current traffic volumes using some priority rules--sample most often the most traveled routes.  But priorities are needed only when resources are scarce; as technology becomes cheaper it would be possible to sample everything all the time (which is almost what the human sensory system does).

Safeway's Self-Checkout and Driverless Cars

Vox has a long piece on the problems with self-checkout.

I have been a fan of self-checkout, which I use regularly at Safeway, but I'm getting less enthusiastic. My local Safeway has probably had self-checkout for 10 years or so.  You'd think that the system would keep working indefinitely but not so.  I suppose it's probably the hardware getting unreliable, but it seems like the software.  It's most noticeable when handling produce--hitting the icons for entering a code or selecting from a screen I often (it seems often) given a system error--needing a sales attendant.

My experience with the self-checkout raises some questions with driverless cars.  My assumption has been that the system will always improve--any problem in the software which turns up will be fixed on all the cars using the software.  But Safeway argues for the law of entropy.  While the software may endure, the underlying hardware and the accessories for input/output won't endure.  They'll degrade. 

I can switch my argument again by pointing to airplanes.  Boeing and Airbus also have a combination of hardware and software which is used over years and years, and they seem to have solved the problem of degrading hardware.  Elon Musk notoriously didn't pay much attention to the experience of established carmakers; I wonder if he will similarly ignore the plane makers.

Toyota--Say It Ain't So--Driverless Cars

Here's a report on a Toyota bigshot's skepticism that we'll get driverless cars any time soon.

I've been driving my leased-Prius 2 for a year, chosen because of its safety features and reasonable price.  The features are good, but not fool proof--I've had a couple close calls, albeit at low speeds so likely the worst result would have been a fender-bender.  But still, I really want improvements in the features, FAST.

Those Stuck-in-the-Past Old Fogeys

Like me, many elderly don't like change.  But it varies, and we can surprise you if it's to our benefit:

"Elderly participants were most excited about the idea of autonomous vehicles, but only 36 percent of young adults were comfortable with the idea of riding in one. "

From the Rural Blog, discussing research into attitudes to self-driving cars.

Drivers Monitoring Autonomous Cars

Two points on autonomous cars:

1. China has just authorized Baidu to run their autonomous cars on the highways.  The piece I saw noted that Chinese roads are more crowded and chaotic than in the US, thereby posing a bigger challenge to the software.  I'd add--doesn't that give them an advantage in development--a higher bar to surmount?
2. AEI notes that humans are poor monitors.  We get distracted and complacent and don't jump into action quickly.  I wonder if it would be possible to include a training module in the software--have the software test the driver by requiring intervention in a situation which is actually safe.  If the driver fails to react timely and correctly, do more testing.  If the driver continues to fail, discontinue the self-driving. 

[Added link]

A Hurdle for Self-Driving Cars

I've been big on self-driving cars in the past, but I just saw a hurdle they'll have difficulty overcoming.

The background: in the DC area we've had snow.  Before the snow VDOT/Fairfax put down brine on the roadways.  So we get our epic 1" snowfall and the conditions of the road (frozen from a long period of freezing temps), the brine and the traffic result in the roads essentially turning white--no way to see lane markings.  Oops.  Presumably they'll give the software enough smarts to identify conditions in which the software can't work.  I hope.

Driverless Cars: For Rural Areas?

NYTimes devotes its magazine this week to the subject of driverless cars.  One prediction of 2-5 years for the most ambitious cars which might work for my case. I'm somewhat dubious over some of the crystal ball gazing, but we'll see. 

My own predictions: driverless vehicles will take off first in niche markets: long distance trucking, Uber/cabs, the elderly.  They won't progress as fast with the mainstream of drivers--people like to control their lives and many will be impatient with the granny-like driving that adherence to rules will foster.   A key will be relative cost:  some of us will pay a premium for driverless cars, others will wait to benefit by lower costs on a per-ride basis.

As time goes by we'll have to change the traffic rules, but that will be difficult with a mixture of vehicles.  

One big hurdle will be rural areas.  At some point, population density will be so low that a driverless Uber/cab service doesn't make sense--it will take too long for the vehicle to get to the user.  For such areas the cost of the driverless car will have be to be less than the cost of the driven car.

The New Car--I

I've owned 4 cars in my life--first a VW Beetle, then Toyota Corollas, the last one a 2006 model.  My driving is getting more questionable these days: more easily distracted and more prone to panic when I get lost are the main symptoms.  But I'm not ready to give up my keys, so early this month I leased a 2017 Prius 2, choosing it mainly on account of the advanced safety features.  It's not the self-driving car I really want, and which I asked (joking) the salesman for, but it's the next best thing, at least in my cost range (not a Tesla 3). 

So wife and I took off for the NY Sheep and Wool Festival in Rhinebeck, NY in the car.  When we got back Monday evening we had something over 1,000 miles on the car, which told us we averaged 61 mpg.  Not bad.