Re: [SystemSafety] Nature article on driverless cars

From: Matthew Squair < >
Date: Mon, 16 Apr 2018 16:30:18 -0700


I’ll stipulate that a car can drive itself when I see one successfully and safely negotiate street traffic in Djibouti, a peak hour temporary roundabout in Marseilles and the California freeway system. :)

So I’ll risk falling foul of one of Clarke’s laws here and predict that much like fusion power true driverless cars will be ‘just five years away’ for a good deal longer. My reason for saying that is that the control problem is just not amendable to simplification. Newton for example was successful in explaining the motions of the planets because he was a genius at making simplifying assumptions that made the problem tractable and then in recognising that the resultant two body answer was good enough, classical engineering has successfully pursued that approach ever since (neural networks need not apply).

Here there’s no ‘two body’ approximation for negotiating your way through traffic, hence the learning networks strategy that’s intended to get around the square law of computation problem. But notice where the current generation of automated vehicles are still being trained (in simple, rule driven predictable environments) and the simplified task (driving and using signals) they’re still being asked to perform. Yes I’m sure that automated bus routes or trams or long haul trucking on the freeway will happen, but these are all very low hanging fruit.

On 17 April 2018 at 2:59:50 am, Bernard Ladkin Peter (ladkin_at_xxxxxx wrote:

On 2018-04-16 17:58 , Peter Bishop wrote:
> Maybe the Nature authors should have concluded that drivers should
> always drive, and automation should only warn, and possibly take a
> safety action like applying the brake if the driver takes no notice.

You know, I don't know (I guess this is already apparent :-( ).

As a cyclist, and non-car-owner for twenty-plus years, I have some history. I have ridden widely for
many years in three countries (UK, California, Germany). I have ridden regularly for most of a year
in two others (France, Switzerland). And sporadically in the Low Countries.

I have only ever collided with cars in Bielefeld, all v. low-speed (3 times, I seem to remember). I
have come within a fraction of a second of very probably being killed by a driver contravening at
least four rules of the road once, also in Bielefeld. I have avoided at least one other
fatally-threatening instance, but with seconds to spare. All of these events would have been avoided
had these vehicles been AVs using the technology which they have today (except for the last, they
were all instances in which I had right of way. The reason the collisions were v. low speed is that
I don't assume my right of way. This has its own issues. The two life-threatening instances were
both cases of significant overspeed of the threatening vehicle, plus other factors).

Generally speaking, if vehicle automation were simply to regulate speed (successfully) to the posted
speed limit, a large proportion of accidents would not occur. This alone would have resolved both of
my two life-threatening instances. A lane-recognition system would have avoided all the others.
Probably in none of these cases did I need to be "seen". In fact, in the two life-threatening cases,
I was obviously seen because the drivers manoeuvred illegally in order to maintain their desired
illegally-rapid progress.

Given my experience, the concept of active drivers with active road-rule limitations has obvious
appeal to me. So does, independent of that, the idea of reliable personal transportation for those
road users who cannot otherwise drive, which was and is going to be most all of us at both ends of
our lives, certainly at the left-hand end. As well as, in between, those overindulgent evenings at
favorite restaurants. And paying less for a taxi.

So, looking at it from the safety point of view, if AVs become universal, am I going to be safer as
a cyclist? It's the plus of speed-control and lane-recognition versus the minus of inscrutable
behaviour.

Looking at it from the social-transportation point of view, it is the plus of widely-increased
independent mobility versus the minus of whatever safety concerns emerge.

Looking at it from the cybersecurity point of view, I think it's living nightmares without end.

PBL Prof. Peter Bernard Ladkin, Bielefeld, Germany MoreInCommon
Je suis Charlie
Tel+msg +49 (0)521 880 7319 www.rvs-bi.de



The System Safety Mailing List
systemsafety_at_xxxxxx


The System Safety Mailing List
systemsafety_at_xxxxxx Received on Tue Apr 17 2018 - 01:30:39 CEST

This archive was generated by hypermail 2.3.0 : Mon Dec 10 2018 - 06:17:09 CET