Tuesday, May 19, 2015

Techy Tuesday - The Self Driving Car Hype

You know, being immersed in the electronics business 24/7/365, I have a tendency to think "everybody knows that" when things about the way our world works come up.  In this case, does everybody know the electronics industry pretty much runs on a hype cycle?  The Gartner Hype Cycle is 21 years old now, and it describes the progress of technologies as well as anything.  There's a good introduction at that link, but the Gartner cycle itself is a detour from where I'm going tonight, except for one thing: autonomous or self-driving cars are running right around the absolute peak of the hype cycle.  Because of that, discount virtually everything you're hearing. 

I've heard pundits say that "a child born today may never learn to drive" (aside from the phenomenon of millennials being less interested in driving).  Both Nissan and GM, as well as some other car makers, say they expect to market self-driving cars by 2020 and some pundits are saying they will be mandatory by 2035. 

A study by a professional research company called Lux Research offers market research to show that's really a lot of hype.  Note this is research they sell, so their survival depends on being right often enough to get repeat business.  Design News extracts some coverage from that report for us. 
”The $102 Billion Opportunity in Partial Automation for Cars" contends that autopilot features in vehicles will grow sharply over the next 15 years, creating huge markets for automotive sensors and software. “Partial autonomy is coming,” Maryanna Saenko, Lux Research Inc. analyst and author of the report, told Design News. “By 2030, it will very likely be common in mid- and high-level cars. But the idea of the car picking you up at your house, driving you anywhere and dropping you off -- that’s still a long way off.”
They include this graphic in the article (not very high resolution) showing that even out at the end of the projections in 2030, fully autonomous cars (darkest blue) don't even show up.  The larger, lighter blue area is "enhanced assist" technologies.
For consumers, the result will be a fast-growing variety of new semi-autonomous features, Saenko told us. “The first step is to have autopilot,” she said. “Not where the car drives endlessly in a straight line, but where it can deal with traffic, merge in and out of lanes, and find the fastest route on the highway.” In subsequent steps, engineers will develop vehicles that can handle suburban and urban driving.
Cruise control speed-regulating devices have been around for decades, and I've had one since 1990.  On long trips they make driving more like just steering; just keep it between the lines.  My first car with cruise control just kept the speed constant.  It also regulated at about 2 mph over the speed you were going when you turned it on.  My next car, in 2004, would change the speed in 1 mph steps.  I'm fairly sure that was digital cruise control.  Autopilot, even if it just kept you between the lines would be convenient, but the ability to do these other tasks she refers to get very close to chauffeur level.
Saenko said technical and regulatory hurdles will complicate the transition from partial autonomy to full autonomy, however. A “likely” scenario described in the report calls for just a handful of fully autonomous cars, operating in highly restricted environments, by 2030. “We have to address how technically difficult it is to have a car drive itself and deal with the errant situations that can happen during driving,” she said. “It’s really about the technology getting competent and reliable enough to deal with all possibilities.”
I should have said, "get very close to chauffeur level, except..."  Except that you are ultimately still in control in the car and will be responsible if it does anything wrong.  That means if your autopilot hits a pedestrian, it's the same as if you were driving.  If you're expecting a day when you can get in the car and read, play games, or conjugate Indian verbs with your girl/boy friend, that's just not happening soon.   Barring new technologies or real breakthroughs.  I actually wrote about this aspect already, so I'll reprise something from just a couple of months ago.
This concerns me because in many ways, driving a car is considerably more complex than flying a plane, yet airplane autopilots get really serious amounts of money dumped into certifying them safe.  Even with that effort and expense, aircraft automatic control systems will still get confused and do the wrong things.  Pilots talk about trying to stay alert at all times should the autopilot hand control back, but one of the disadvantages of the modern autopilots is that they're so good pilots get out of practice flying.   Will self driving cars caught in a critical moment hand control off to someone reading a book on their tablet, listening to music or otherwise engaged with people in the car? 
Driving a car in a city environment is more complex than flying a plane.  Nobody is standing around on a cloud and steps out in front of a plane.  There are no planes sitting at a corner that decide to dart in front of you.  There are no bicycles going a quarter of your speed, or old farm tractors going even slower.   The fact the aircraft on similar routes maintain separation by flying similar speeds helps reduce the chance of an overtaking accident.  Yet with all these things going for them aircraft autopilots still get confused.  We're going to have to go through some years of weeding those failures out of autonomous cars. 


  1. Truly driverless vehicles are a LONG way off. They will arrive when we reach that feared point when computers and programs become self aware, sentient. Until then a 'driverless' car is only as good as the sensors it uses and the programming it relies on. And making a car ALMOST driverless yet keeping the human in the loop as the 'responsible' part of the machine is stupid. People need to practice driving just like anything else. If not they are no good at it and won't make the correct decisions when the machine FUBAR's things.

    Unless and until we see significant improvement in a number of technology's driverless vehicles won't be practical and won't be common. And by the time that level of technology arrives we may just be too busy fighting
    SkyNet to worry about cars.

  2. I suspect what we'll see is an increase in number and quality of semi-autonomous safety and convenience features - collision avoidance, lane tracking, better anti-lock braking, "training", etc. - all of which will get steadily better but will still require a conscious driver. Improvements and redundancy will continue until the threshold of true autonomous operation is reached, which will be a great deal more difficult to achieve than most people realize.

    A couple decades back I had to disabuse management of automating a particular process; essentially, they wanted to automate a complex analog process, and the number of discrete digital points necessary was orders of magnitude greater than the technology could handle in either number or time, not to mention the fragility of the necessary software. That also largely describes the simple-sounding task of "driving a car."

    There's a desire within industry in general and the tech industry in particular to reduce the cost of the "necessary human" by increasing automated operation and control, allowing both an increase in human span of control and a substantial reduction in the human skill level required, meaning "cheaper people" with what they hope is an end point of "no people."

    The trucking industry is playing with "training" to electroncally couple multiple trucks to allow speed increases and drag reductions. It's a workable concept, and if successfully implemented will bleed over into automobiles, but it's quite far from autonomous operation and still requires a driver. I'll predict some really stupendous accidents with it as they cut corners to improve the ROI.

  3. CMU (Carnegie Mellon University) has had a self driving car operational since 1991 - and you don't hear them hyping it!
    You are right on - the reason unmanned aircraft are around is the relative ease of designing an autopilot; an unmanned military ground vehicle is a whole different can of worms - dealing with traffic laws, pedestrians, and other vehicles is a huge step in both ability and liability.
    I have heard that California issued driver's licenses to 3 car companies for self driving cars; I would be interested in what the conditions of operations were for them - I would not be surprised to find out they were limited to lightly populated areas in daylight and good weather conditions.
    I will be very concerned if Google actually fields a driverless car - software companies, particularly internet companies, have a history of beta programs and never really finishing the job; that is unacceptable in the hardware world and particularly in safety and life critical applications.

  4. I can see automating the freeways in between large cities. You merge in to traffic and press a button...

    When you approach an area where auto control is not available, a very loud horn would sound telling you to resume control in 5, 4, 3, 2..

    I wouldn't mind at all leaving my house around 8PM, getting on the freeway, and letting the car drive me to wherever while I sleep. Maybe not in my lifetime...but I'd much rather do that than fly in a sardine can.

  5. While there's a huge hype for the self driving car, I'm pretty sure that our kids will want to learn how to drive. Just because they don't have to, doesn't mean they won't WANT to, and I certainly hope they do!