F. Fishkin,
Aug 3, with
Michael Sena
about what to
expect in
2030 Listen
to Hmmmm... We are still at
the very
beginning.
Alain
M. Harris, Aug 4, "While automakers focus on defending the systems in their cars against hackers, there may be other ways for the malicious to mess with self-driving cars. Security researchers at the University of Washington have shown they can get computer vision systems to misidentify road signs using nothing more than stickers made on a home printer. UW computer-security researcher Yoshi Kohno described an attack algorithm that uses printed images stuck on road signs. These images confuse the cameras on which most self-driving vehicles rely...." Read more Hmmmm... Well, that's not quite right. It confused "one road sign classifier... consisting of three convolution layers followed by a fully connected layer trained over 100 epochs using the LISA data set". So, this result is far from "most" of anything. Given that the original model was 'only' 91% correct, its testing set identified perfectly good examples of miss-classifications.
The paper "Robust
Physical-World
Attacks
on
Machine
Learning
Models" by
I. Evitimov is
interesting
because it
lays out
approaches to
generating
robust
realistic
adversarial
examples of
real objects
but the title
should, at
best, start
with the word
'Toward'.
Let's wait to
see if the
findings/results
are
reproducible
for this
specific
classifier
before we jump
to
'robustness'
and 'all' ML
models. Alain
J. Laughlin, July 23 "...Compared with 2013, the last full year SEPTA operated without ride sharing in the city, the 2016 ridership loss is even more significant. Last year’s more than 161 million ride trips on both buses and trolleys was about 14 million shy of the ridership three years earlier..." Read more Hmmmm... What are the implications on SEPTA when autonomousTaxis come to Philly? Alain
A. Hwang, July 27. "...The EZ10 driverless vehicle will run on a bus lane from 1:00-4:00 am every day during August 1-5..." Read more Hmmmm... Bus lanes. OK. It is another attempt at a start, which is good. Alain
A. Liptak,
Aug 6,
"...They used
a normal
production
vehicle
outfitted with
low rolling
resistance
tires with the
air
conditioning
turned off.
Five drivers
completed the
task in
southern Italy
in 29 hours,
and noted that
the drive was
made easier
with the car’s
autopilot
turned on,
which “helped
us keep a
constant speed
in the middle
of the
lane.”..." Read
more Hmmmm... And they started at the
top of the
Alps. ;-)
Alain
http://orfe.princeton.edu/~alaink/SmartDrivingCars/Papers/
M. Sena,
Vol 4, issue
9,
"UNCERTAINTY
IS TROUBLING
for
businesses,
individuals
and
governments....
In one way or
another, all
businesses,
including and
especially
transport, are
completely
reliant on
four macro
factors:.. I'd add one more: where are children learn and play
.... A United
Nations study
projects world
population to
reach 8.5
billion by
2030, up from
7.5 billion
today, driven
by growth in
developing
countries...India
will have
traded places
with China as
the world’s
most populous
country in
around seven
years...So the
large bulk of
those
additional one
billion
inhabitants of
the planet by
2030 will be
looking for
places to live
in Mumbai, not
in Madrid. The
takeaway from
this is that
the so-called
‘developed’
countries,
with a few
notable
exceptions,
are either
losing
population due
to not
producing
enough
children or
seeing their
populations
staying
basically
stable. In
2030, Tokyo
will still be
the most
populated city
with an
estimated
population of
37.2 million.
Delhi will be
in second
place with
36.1 million,
up from 3.5
million in
1970! (How has
it coped?)
Shanghai will
be in third
place and New
York/Newark
will have
dropped off
the top ten
list. But what
will it be
like to live
in these
cities? The
Economist
Intelligence
Unit ranks
cities as the
most and least
liveable. ...
It ranked
Melbourne,
Australia as
number one,
...
Melbourne’s
density is 460
persons per
km2 compared
to 6,158/km2
for Tokyo and
2,059/km2 for
Shanghai....
None of the
most liveable
cities is
among the top
ten places
where venture
capitalists
have been
placing their
money bets
during the
past
year....These
four city
regions are
ranked below
30th place on
the EIU
Liveability
Index. In
other words,
they may be
successful,
but not that
liveable...
(in US) 50%
live in rural
or less urban
areas
occupying more
than 90% of
the land area.
Is there any
wonder why
over 50% of
the vehicles
sold in the
U.S. are not
passenger cars
but SUVs and
pick-up
trucks?...If
everyone who
lived in the
dense urban
areas stopped
buying cars,
there would
still be over
50% of the
population who
would continue
to be car
purchasers.
Can we conclude from this that the exodus from city regions to the suburbs of both jobs and families has now stopped and central cities once again will be where people live and work? No, not unless people will be willing to give up everything they have come to value in terms of living standards and will accept being packed into sardine can-sized apartments stacked a mile high.... Living in a central city in the most desirable neighborhoods will continue to be the privilege of the wealthy and very wealthy...They also have homes and dachas in the Hamptons, Vinyard and Vermont, else they couldn't stand it. ... When younger people build families and need more space, preferably with a yard, and that space is too expensive in the city, they find it further out...Visions of young professionals dashing around in robotic cars gobbling up mobility as a service are, to put it kindly, a bit fanciful. Read more Hmmmm...I love it!! So many good one-liners. Alain
D. Hall,
Apr 17, "In
the race to
the autonomous
revolution,
developers
have realized
there aren’t
enough hours
in a day to
clock the
real-world
miles needed
to teach cars
how to drive
themselves.
Which is why
Grand Theft
Auto V is in
the mix.
The
blockbuster
video game is
one of the
simulation
platforms
researchers
and engineers
increasingly
rely on to
test and train
the machines
being primed
to take
control of the
family sedan.
Companies from
Ford Motor Co.
to Alphabet
Inc.’s Waymo
may boast
about putting
no-hands
models on the
market in
three years,
but there’s a
lot still to
learn about
drilling
algorithms in
how to respond
when, say, a
mattress falls
off a truck on
the
freeway....The
idea isn’t
that the
highways and
byways of the
fictional city
of Los Santos
would ever be
a substitute
for bona fide
asphalt. But
the game “is
the richest
virtual
environment
that we could
extract data
from,”
said Alain
Kornhauser..."
Read
More Hmmmm... Well...we have a slightly different
view of
history wrt to
GTA5. The
'Alain view'
is that Chenyi
Chen*16
independently
started
investigating
the use of
virtual
environments
as a source of
Image -
Affordances
data sets to
use as the
training sets
in a 'Direct
Perception'
approach to
creating a
self-driving
algorithm.
Images of the
road ahead are
converted into
the
instantaneous
geometry that
is implied by
those image.
An optimal
controller
then
determines the
the steering,
brake and
throttle
values to best
drive the
car. The
critical
element in
that process
are the Image - Affordances data
sets which
need to be
pristine.
Chenyi
demonstrated
in his PhD
dissertation
, summarized
in the ICCV2015
paper,
that by using
the pristine
Image -
Affordances
data sets from
an open-source
game TORCS
one could have
a virtual car
drive a
virtual race
course without
crashing.
More
importantly,
when tested on
images from
real driving
situations,
the computed
affordances
were close to
correct.
This encouraged us to look for more appropriate
virtual
environments.
For many
reasons,
including:
"wouldn't it
be amazing if
'Grand Theft
Auto 5'
actually
generated some
positive
'redeeming
social value'
by
contributing
to the
development of
algorithms
that actually
made cars
safer; saving
grief,
injuries and
lives".
Consequently,
in the Fall of
2015, Artur
Filipowicz'17
began to
investigate
using GTA5 to
train
Convolutional
Neural
Networks to
perform some
of the Direct
Perception
aspects of
automated
driving. With
Jeremiah Liu,
he continued
his efforts in
this direction
last summer
which were presented
at TRB in
January.
Yesterday, he
and Nyan
Bhat'17
turned in
their Senior
Theses focused
on this topic.
A. Kornhauser, Jan 14, "Orf467F16 Final Project Symposium quantifying implications of such a Nation-wide mobility system on Average Vehicle Occupancy (AVO), energy, environment and congestion, including estimates of fleet size, needed empty vehicle repositioning, and ridership implications on existing rail transit systems (west, east, NYC) and Amtrak of a system that would efficiently and effectively perform their '1st mile'/'last-mile' mobility needs. Read more Hmmm... Now linked are 1st Drafts of the chapters and the powerPoint summaries of these elements. Final Report should be available by early February. The major finding is, nationwide there exists sufficient casual ridesharing potential that a well--managed Nationwide Fleet of about 30M aTaxis (in conjunction with the existing air, Amtrak and Urban fixed-rail systems) could serve the vehicular mobility needs of the whole nation with VMT 40% less than today's automobiles while providing a Level-of-Service (LoS) largely equivalent and in many ways superior than is delivered by the personal automobile today. Also interesting are the findings as to the substantial increased patronage opportunities available to Amtrak and each of the fixed rail transit systems around the country because the aTaxis solve the '1st and last mile' problem. While all of this is extremely good news, the challenging news is that since all of these fixed rail systems currently lose money on each passenger served, the additional patronage would likely mean that they'll lose even more money in the future. :-( Alain
September
2016,
"Executive
Summary...For
DOT, the
excitement
around highly
automated
vehicles
(HAVs) starts
with safety.
(p5)
...The
development of
advanced
automated
vehicle safety
technologies,
including
fully
self-driving
cars, may
prove to be
the greatest
personal
transportation
revolution
since the
popularization
of the
personal
automobile
nearly a
century ago.
(p5)
...The benefits don’t stop with safety. Innovations have the potential to transform personal mobility and open doors to people and communities. (p5)
...The remarkable speed with which increasingly complex HAVs are evolving challenges DOT to take new approaches that ensure these technologies are safely introduced (i.e., do not introduce significant new safety risks), provide safety benefits today, and achieve their full safety potential in the future. (p6) Hmmm...Fantastic statements and I appreciate that the fundamental basis and motivator is SAFETY. We all have recognized safety as a necessary condition that must be satisfied if this technology is to be successful. (unfortunately it is not a sufficient condition, (in a pure math context)). This policy statement appropriately reaffirms this necessary condition. Alain
"...we
divide the
task of
facilitating
the safe
introduction
and deployment
(...defines
“deployment”
as the
operation of
an HAV by
members of the
public who are
not the
employees or
agents of the
designer,
developer, or
manufacturer
of that HAV.)
of HAVs into
four
sections:(p6)
Hmmm...Perfect!
Alain
"...2.
Model State
Policy (p7)
The Model
State Policy
confirms that
States retain
their
traditional
responsibilities...but... The shared
objective is
to ensure the
establishment
of a
consistent
national
framework
rather than a
patchwork of
incompatible
laws..." Hmmm... Well done. Alain
"...3. NHTSA Current Regulatory Tools (p7) ... This document provides instructions, practical guidance, and assistance to entities seeking to employ those tools. Furthermore, NHTSA has streamlined its review process and is committing to..." Hmmm... Excellent. Alain
"...4. New Tools and Authorities (p7)...The speed with which HAVs are advancing, combined with the complexity and novelty of these innovations, threatens to outpace the Agency’s conventional regulatory processes and capabilities. This challenge requires DOT to examine whether the way DOT has addressed safety for the last 50 years should be expanded to realize the safety potential of automated vehicles over the next 50 years. Therefore, this section identifies potential new tools, authorities and regulatory structures that could aid the safe and appropriately expeditious deployment of new technologies by enabling the Agency to be more nimble and flexible (p8)..." Hmmm... Yes. Alain
"...I.
Vehicle
Performance
Guidance for
Automated
Vehicles
(p11) A.
Guidance: if a
vehicle is
compliant
within the
existing FMVSS
regulatory
framework and
maintains a
conventional
vehicle
design, there
is currently
no specific
federal legal
barrier to an
HAV being
offered for
sale.(footnote
7) However,
manufacturers
and other
entities
designing new
automated
vehicle
systems
are subject to
NHTSA’s
defects,
recall and
enforcement
authority.
(footnote 8)
.
and the "15
Cross-cutting
Areas of
Guidance"
p17)
In sum this is a very good document and displays just
how far DoT
policy has
come from
promoting v2v,
DSRC and
centralized
control,
"connected",
focus to
creating an
environment
focused on
individual
vehicles that
responsibly
take care of
themselves.
Kudos to
Secretary Foxx
for this 180
degree policy
turn focused
on safety.
Once done
correctly, the
HAV will yield
the early
safety
benefits that
will stimulate
continued
improvements
that, in turn,
will yield the
great
mobility,
environmental
and
quality-of-life
benefits
afforded by
driverless
mobility.
What are not addressed are commercial trucking and buses/mass transit. NHTSA is auto focused, so maybe FMCSA is preparing similar guidelines. FTA (Federal Transit Administration) seems nowhere in sight. Alain
Hmmm...What we know now (and don't know):