Friday, September 9, 2016

Self-Driving Cars Since The Famous Tesla Crash: A High-Speed Round-Up

As, per John Markoff of The New York Times, “in and around Silicon Valley, at least 19 commercial self-driving efforts are underway,” the topic of driverless vehicles continues to set publication proliferation records among jobs-related subjects.  I am looking at a stack of 29 articles from no fewer than 12 different publications, all within the past 73 days.  I won’t try to review all of them, which would be a waste of space for several reasons, but will summarize what has happened here, sorted into four major themes, since word of the first fatal self-driving mishap broke on June 30.

This crash took place May 7th in Williston, Florida, when a Tesla driver activated the car’s Autopilot feature, completely ignored what was happening on the road, and then died when his vehicle ran full speed into the side of an 18-wheel truck.  After almost three months of investigations (we’ll need to get faster here), and initially saying that the accident happened because the product did not correctly interpret “the white side of the tractor-trailer against a brightly lit sky,” the Tesla company determined that the problem was not that Autopilot failed to perceive that there was a large metal mass straight ahead, but that its braking system did not then stop the car.  In the meantime, Consumer Reports implored the company to block use of this feature while the drivers’ hands were off the wheel. 

A few observations here.  Apparently Autopilot, as well as being named aggressively enough to encourage drivers to, in the findings of another article, play Jenga, watch Harry Potter, and actually sleep, is in fact a beta release, meaning that it has not been fully tested.  If the linkage between its autonomous driving capability and its brakes is insufficient, it is in effect little more than cruise control, around since at least the 1960s and nothing any prudent driver would ever trust with full vehicle operation.  It is not, as I defined the levels in my July 1 post, a true Plane 2 feature, as drivers cannot disengage while it is operating.  Whether the actions above were reckless, appropriate given the name of the feature, or somewhere in between, and even though the cars’ manuals gave warnings, it is clear that Tesla dropped the ball here.  They will probably lose large lawsuits over the product, and will be required to back it off in some ways.  The broader ongoing issue will be to avoid excessively slowing down availability of automated driving products.  There will be more mishaps – in fact, on July 1 another Tesla crashed, rolling over on the Pennsylvania Turnpike without fatal injuries – and their liability here does not mean the maximum acceptable rate of accidents should be zero.

The second subtopic here is about company alliances and consortiums, which I predicted July 1 would ultimately produce these vehicles.  Later that same date, German automaker BMS AG, American chipmaker Intel, and self-driving Israeli software company Mobileye NV announced that they, together, expected to be producing self-driving cars of some autonomy level by 2021.  About three weeks ago, American ridesharing company Uber and Swedish car manufacturer Volvo announced combining to offer automated taxis, which will be on Pittsburgh roads this fall.  In good deference to their necessarily incomplete quality, they will be free, and at Plane 2 or Plane 3 (a driver still inside, but rarely taking the controls).  How this aggressive and possibly premature but fascinating experiment works, or fails, will tell us a lot about what we need to stay on my projected track of 20% Plane 2 United States vehicles by 2021 and 20% Plane 3 ones by 2025.  Uber has also acquired American driverless truck technology maker Otto, which gives it two consortium pieces on that front in-house.  Ford, now with potent production plans of its own, is working with Israeli computer-vision company Saips, has bought 75% of American lidar (radar based on laser beams – expect to see this term more and more) sensor manufacturer Velodyne, and put undisclosed amounts of money into American digital chart firm Civil Maps and American machine-vision concern Nirenberg Neuroscience.

The third theme concerns announced availability dates for various vehicle automation products.  A July 4 article said that Google “hopes” to market what might be called self-driving bumper cars, with 25-mile-per-hour maximums, “a heavy later of foam” on their fronts, and plastic windshields, in 2019.  Those would go all the way to Plane 5, or vehicles without direct human control; although I projected they would not reach 20% of American rolling stock until 2030, we could still reasonably have 0.1% in three years.  General Motors now plans on a 2017 offering of Super Cruise, an apparent Tesla Autopilot competitor, designed as a true Plane 2 product but functional only on highways which the company has mapped in detail.  Ford announced on August 16th offering large numbers of fully driverless cars, at Plane 5 or at least Plane 4 (remote human supervision of vehicles or groups of vehicles; my prediction 20% in USA by 2028), through a ridesharing company to be named by 2021.  And in four miles of Singapore streets, American company nuTonomy has zoomed ahead of Uber and Volvo by offering free experimental semi-driverless taxi service now.

The fourth article concentration is on other updates to issues I raised in my November 20th post.  A July 7th piece proposed that, like 16-year-old humans, new driverless technology pieces should be licensed – a sensible idea, especially if the turnaround time is much shorter than those government often perpetrates.  A second, on July 8th, reinforced the special need for road and other infrastructure improvement.  A third, also out July 7th, again raised the behavior issue, with Plane 2 drivers averaging 17 seconds, or a quarter of a mile at only 53 MPH, to retake control after the vehicle’s command to do so.  An August 18th article addressed one I haven’t seen since bringing it up in November, the threat of hacking, with a graphic example of a researcher stopping a self-driving car by issuing commands on his laptop.


As far as jobs go, my July projections on those, as with the percentage implementation dates, are still current.  Uber seems to be insisting that driverless vehicles will slash the number of privately owned cars, which seems unlikely in small towns and low-density suburbs and absurd in rural areas.  I’m still expecting 20% of American vehicles at Plane 2 by 2021, privately owned or not, with the world’s self-driving leaders there by 2018.  And eventually our terminology will change.  Per Ben Zimmer on August 26th, we may come to know of self-driving cars as just “cars,” and the ones we have now as… “meatmobiles.”  I leave it to you to propose other names for what were once known as “horseless carriages” – in the meantime, enjoy watching the progress of what might, arguably of course, be the most exciting, and most promising, area of technical growth we have seen since Apollo.

No comments:

Post a Comment