Parts 1 and 2 of this blog set a contrarian tone, discussed surveys of public opinion, road testing and results-and some practical problems for the autonomous vehicle (AV).  There is much hype going around. This concluding part will look at government involvement, ownership, some operational considerations and finally when will we know they’re safe.


Traditionally the Federal Government has taken on overall safety requirements (crash protection, safety belts) and the states have dealt with more local issues (operating licensing, emissions testing, insurance). In 2017 the Department of Transportation National Highway Traffic Safety Administration (NHTSA) released Automated Driving Systems 2.0. It provides a “nonregulatory approach to automated vehicle technology safety” for what the document calls the Automated Driving System (ADS).

The Federal Government is clearly taking a lassez faire approach to the new technology Elaine Chao Secretary of Transportation voicing “We’re not in the business of picking winners or losers. The market will decide what the most effective solution is.”  There goes the dream of cars talking to each other. Who will dictate the communication protocol?  This approach ignores the fact that the Code of Federal Regulations (as you can guess a list of all Federal regulations) includes the Federal Motor Vehicle Safety Standards (FMVSS) which cover things like side crash protection, air bags, child restraints and door locks.  Are we going to ignore these?  Remember, as of 2016 there are 263 million vehicles registered in the US and the vast majority of them will continue to be driven by humans.


There won’t be any level 5 AVs in show rooms for the foreseeable future.  The Economist Magazine says that the assumption that they will resemble cars, that you will be able to buy these anywhere and that they will work everywhere in all road conditions may be wrong. Given the current testing it is more likely that the first commercially available models will be pod-like for urban rise sharing.  Ford is pledging its first models to fleet sales (car rental companies are looking at them) and of course Uber and Lyft are working on their own models.  At this point the cost of the sensors is greater than the actual car. But will level 5 AV be commercially desirable?  I refer again to the apparent limited market appeal in Part 1


While fatalities in urban and rural areas are about equal in 2016, there are 2.5 deaths per 100 million miles driven in rural areas versus 0.8 in urban areas. Thirty-seven percent of the fatalities are in passenger cars and minivans. Fifty five percent are single car and forty five percent multi-cars.


I am old enough to remember when televisions first showed up in homes.  Because there was a dearth of programming, air time was filled with things like the AAA safe driving movie which among other things showed the risk of taking curves too fast by showing a model car veering of a  speeding turntable and saying that said you needed one car length  between cars for every 10 mph.  Of course there weren’t a lot of cars on the highway and this was a decade before the Interstate highway system.  Later it was the 2-second rule, you had to make sure there two seconds travel time between cars.  Fine if you’re on a 2 lane highway with telephone poles but more difficult on freeways.

It’s my observation that experienced highway drivers have a sense of vehicle spacing and with anti-lock braking systems in just about all cars today stopping short on dry pavement is much more effective.  It’s just that people get stupid or impatient driving in inclement weather.  Even with the faster response time of interconnected cars there’s the physics of stopping which is much longer than 2 seconds. I don’t think you’ll cars spaced like this traveling at 50 mph.  Here’s a table of how much is traveled during the reaction time (faster in AV) and the actual breaking time (based on physics) at at various speeds

Car spacing


Here is what some of the AVs will look like.

This is not a safe car


If 45 percent of fatalities are multi-car don’t we need something stronger?  Maybe these will be limited to dense urban areas but there are multi-car fatalities there.

In California’s Senator Diane Feinstein’s letter of March 14, 2018  to John Thune Chairman Committee on Commerce, Science, and Transportation she wrote “Self-driving cars should be no more likely to crash than cars currently do, and should provide no less protection to occupants or pedestrians in the event of a crash (as defined in the Code of Federal Regulations)

The organization that is in charge of car safety testing nationally is the Insurance Institute for Highway Safety (IIHS). It issues the annual crash ratings and makes the iconic videos of cars crashing into barriers with dummies inside. Over time they have established among other things that in a two car crash bigger cars survive better than smaller ones.  Also that in 2016 although 19 percent of people in the U.S. live in rural areas and 30 percent of the vehicle miles traveled occur in rural areas, 2 more than half of crash deaths occur there.

Speaking of all safety standards what about seat belts and air bags? I did see a video of someone getting in Google’s “bug” car and putting on seat belts.  Since AVs will shareg the roads forever with SUVs and tractor trailers we don’t want the passengers bouncing around in an accident.

There is discussion about remote monitoring of AVs similar to drone piloting but these work on a one-to-one basis.  Air traffic controller work with multiple planes butt these have human pilots on board and usually don’t need  decisions  in seconds.   Here is a picture from Phantom, a company already doing remote AV control.  It looks like they’d be limited to one vehicle at a time.

Phantom Car Monitor


The RAND Corporation (for Research and Development) a US based non-profit think tank was started by Douglass Aircraft Company in 1948 to provide research services to the United States Armed Forces, funded by the US government and foundations to apply and other public and private clients to address a wide range of interdisciplinary issues from defense to healthcare.

In 2016 they published a study Driving to Safety using their analytical prowess to answer the above question. Americans drive nearly 3.1 trillion miles per year and the 32,719 fatalities in 2013 correspond to a failure rate of 1.09 fatalities per 100 million miles. They conclude that test-driving is not a feasible way to prove the performance of AVs for extensive deployment and they ask “how many miles would autonomous vehicles have to be driven without failure below some benchmark:

They apply existing failure rate statistics with complex formulas (included but above my ability). What they find is that to match the 1.09 per 100 million miles rate the car would have be driven 275 million failure-free miles. With a fleet of 100 autonomous vehicles being test-driven 24 hours a day, 365 days a year at an average speed of 25 miles per hour, this would take about 12.5 years.

Here’s a table showing the testing time for achieving various performance goals.  It’s clearly depressing.


Putting the cars on the road for the public before knowing how safe they are is risky but states and the Federal governments have said they want to apply a light hand to regulation.. In the early days of jet airlines where crashes were more frequent but people still flew and new technology more than legislation increased the level of safety

RAND published a follow-up in 2017 “The Enemy of Good: Estimating the Cost of Waiting for Nearly Perfect Automated Vehicles”  The assumption is that the state of the art gets better as more miles are driven.  RAND used their advanced analytical method this is extremely complex and require several pages to describe in words. They assume the release point is when the AVs are proved to be 10 percent safer than human driver but this requires millions of miles driven. In ay event they assume that the state of the art will improve as time goes by and calculate that at release at this point will save more lives than waiting for 90 percent improvement.

It seems sensible that AVs should be allowed on U.S. roads once they are deemed safe as human drivers and that the state of the art will improve.  RAND recognizes that there will be a problem with social acceptance but it should be done on a Federal level like all the safety standards have been. Seat belts were required in 1967 but it took until 1982 that there were enough cars on the road that the fatality rate dropped significantly.  Waiting for perfect will lose lives.

While the changes to urban form from the advent of AV may be decades away because of the rate of deployment people are talking about like they’ll happen in five years so this will be the focus of my next blog.