Tuesday, July 26, 2016

Again: Self driving cars

This is my third commentary about self driving cars, see here and here for my previous write-ups. Since then there has been several autopilot accidents, and at least one death (the driver).  Now the issues are interesting, insofar as the driver who died (and had been a well known early adopter of Tesla autopilot feature) was killed because the software could not recognize what it saw (the side of a truck).  The single most important need for generation 4 or level 4 autonomous driving (where there is no need for an operator) is that the computer needs to recognize what it sees.  

Five years ago, the hardware needed to understand the real world was almost a super computer, today its a chip!  NVIDIA created a $200 chip that is actually better at recognizing the real world than humans are able to do, a major shift in the ability of machines to recognize the real world -- and adapt.

An article by Chris Dixon here says about the same thing I wrote a few months ago.  I think where is argument is the strongest is autonomous driving for long distance trucking.  Two reasons:  (a) it is incredibly borring task that requires very little mental agility (but a lot of concentration), (b) there is a real shortage of drivers -- its long hours and meagre pay are not incentives in an economy where unemployment is low (around 5%).   The drivers that will remain will be better paid, because they will do more difficult drives that the autodrive system cannot perform well.  Where in fact, humans are the necessary ingredient.

As for the rest of the article, I though that Dixon was a bit pie in the sky.  First off, those who do not drive automatic cars will still need parking for years.  Granted congestion will drop, and if the bulk of these autodrive cars are electric then there should be a huge impact on city pollution. 

What are the missing ingredients for utopia; in no particular order we have: 
  • (a) legislation -- but I suspect it will be a competition from city and states to be "the first", with certain car centric states being the last, 
  • (b) level 4 autodrive vehicles; we are still a few years away from that; currently only about 1 million miles have been driven by autonomous vehicles, while this may seem like a lot, its not,
  • (c) infrastructure  is still far from adequate; there are far too few charging stations in urban environment to allow this "great leap forward" to occur.   In addition the implication for electricity demand is enormous and largely unaddressed
In fact, 90% of what is needed to make this reality happen is still missing...so the early adopters love the the technology, and it works better than advertised (remember google glasses, anyone?).  Of course there have been some real problems, and also so BS, from people who still don't understand how much information is available to Tesla after an accident...in one case the guy basically lied he was driving the car manually, but tried to blame the autopilot feature...dumbass!


Post a Comment

Subscribe to Post Comments [Atom]

Links to this post:

Create a Link

<< Home