Avoiding Accidents
The death of
a young girl outside a school the other day prompted some thoughts about
autonomous vehicles (the accident was caused by a human driver). Will the autonomous
vehicles receive updates over the air, so they are all exactly the same. Or will
they be capable of some learning so they can adapt to local conditions? The possible
combination of:
Heavy rain
Giant hail
Snow
Black ice
High winds
Flash flooding
Narrow, winding streets in the old parts of a city
A new suburb, with roads too narrow for cars parked on the side of the road and
two-way traffic.
Temporary road closures and diversions (GPS maps could handle these)
Driving with poor visibility, or at night
Driving with a damaged system
How much
effort do we put into teaching people to drive (in NSW)?
We teach
them English, so they can read the Road Rules and signs on the road.
They can
apply for a Learner Driver Licence at 16 years of age, and must accumulate 120
hours of logged driving before proceeding to a Provisional Licence.
Humans have
a severe weakness – how many things they can think about at once. They have
demonstrated that driving carefully is within their range, except in the most
arduous conditions, or with a mobile phone held to their ear.
Excerpts from Transport for NSW Road User Handbook
Give way to pedestrians
As a driver, you must give way to
pedestrians:
·
………..
·
when entering or leaving a driveway
What to do
if you are leaving a driveway, and there are walls or fences, so all that is
visible of the footpath is the driveway? Sound your horn? It could have been a school for
deaf children.
Children
Children have not developed
the skills to understand and react to danger. They’re still learning where to
cross safely, and they can find it hard to judge the speed and distance of
vehicles. This means they can act unpredictably around traffic.
Take extra care near:
·
children playing,
walking or riding bikes near the edge of the road
·
schools,
particularly when children are arriving or leaving
·
school buses or
school bus zones where children may be getting on or off the bus.
This is
the crux of the problem. A teenager of 16 will (likely only just) understand all
the references in the road rules: child, driveway, school, bus, developed, unpredictably.
An autonomous vehicle has no structure which allows it to comprehend what is
said in the road rule. Yes, it can be programmed, but that will miss all the
complex relationships linking “understand”, “danger”, “speed”, “visibility”. If children are not able to do these things, the autonomous vehicle must do it for
them.
We
strongly recommend that autonomous vehicles on our roads have sufficient
English to read and understand the Road Rules.
Of course
you would say that – you sell Semantic AI!
OK, but how
else do you make the machine that is the car understand the rules? The machine
has to create generalised objects (children are not short adults) with complex
attributes. Machine Learning? It would have to knock down thousands of children
before it learnt that it was a bad idea. Instead, the machine has to anticipate, be alert
to, and guard against, the possibility. This is not how autonomous vehicle software currently
works.
We have an
Unconscious Mind, which ic continually scanning for problems:
The Road
Rules again
Scanning
Scanning is essential for
safe driving. Scanning is keeping your eyes moving, checking in one area for a
couple of seconds and then moving your eye to another area.
When scanning look:
- · in the distance
- ·
at the road surface
- ·
to your left and
right
- ·
regularly at your
mirrors and instruments.
![]() |
| Orion Semantic AI |


Comments
Post a Comment