Re: The coming third wave of innovation is signal processing

Traveler <traveler@xxxxxxxxxx> wrote:
> Here's an article at embedded.com that's related to recent discussions
> about DARPA's Grand Challenge.
>
> http://www.embedded.com/showArticle.jhtml?articleID=171000445
>
> A couple of excerpts from the article:
>
> A new wave of DSP-based innovation is just around the corner,
> according to Gene Frantz, a principal Fellow at Texas Instruments
> Inc.
>
> ------
>
> In describing the areas where the next wave of DSP innovations
> would most likely occur, Frantz began with transportation,
> highlighting the concept of an autonomous automobile requiring no
> driver, which he said could happen in the next 20 years. As
> evidence of developing technology in this area, Frantz referenced
> the Defense Advanced Research Projects Agency (DARPA)'s second
> Grand Challenge autonomous-vehicle competition, which is
> scheduled to hold a qualification event next week in Fontana,
> Calif.
>
> Dismissing automobile manufacturers who routinely tell him that
> the autonomous car will never happen, Frantz said, "Someday we
> will look back and say, 'How in the world could you ever have had
> a car that wasn't self-driving?"

I'm sure that will happen at some point, but I'm not so sure it will happen
in the next 20 years. I think that's because the current roads and safety
systems (road signs/stop lights/emergency vehicle warning lights), are all
optimized under the assumption of full human intelligence combined with
full human knowledge is controlling the cars and trucks. We might actually
have something like that working in machines in 20 years, but it will most
likely still be too expensive to justify putting into a car (it might cost
$10 million), and more important, society won't be ready to trust the roads
to these machines that soon in any case.

More likely, instead of building full human intelligence and knowledge into
a car, we will just build intelligence which is optimized for the task of
driving, but doesn't understand anything else about the world. It will be
very good at driving and not hitting things (ultra smart cruse control),
but won't have the high level planning and reasoning skills we find in
humans. It will probably be linked into land based systems which track and
report traffic and road conditions to allow the car to always find the most
optimal route and give the passengers very accurate ETA reports.

For route planning, we might end up making a phone call (on-star-like) to a
central system which has more human-like intelligence which we can talk to
about where we want to go and it can find the destination and transfer the
destination and route to the not-so-smart car.

However, because of public pressure, these auto-drive systems will probably
only be allowed on controlled access roads (like the freeways). So, the
conversion of the infrastructure of all the streets from the current
human-drive, to this future where auto-drive is the standard is going to
require a huge investment in money and time. It's likely to take 50 years
even after we have the technology to the level that people are willing to
trust it and get the cost down to the point they are willing to pay for it.

Another reason it won't happen fast is because when we are in our car,
we've got little else to do anyway, so it's not like an auto-drive system
is really worth that much to us. It's not something that can save us a lot
of money.

Where it can save money is in the trucking and taxi business. But it won't
be used there until the system is safe enough and cheap enough, that it
actually works better, and costs less than a taxi driver or truck driver.
And once again, these automated trucks and cabs will only be allowed on
roads dedicated to them - so the conversion from the current system to the
automated system is just going to many decades.

> There is more interesting stuff at the provided URL. Please read the
> article and let me know what you think of the near-term feasibility of
> some of Frantz's ideas.

I think security could be a big area of technology growth with the type of
ideas he was getting at. Though I don't know how much will really be DSP
based.

For example, he talked about using face recognition to allow security
systems to spot danger without building security systems that simply
recorded our actions and invaded our privacy. I think in the end, any scan
device will end up recording as well as analyzing data so I don't think the
privacy issue will be addressed with the new technology, but I do think
there's a lot of stuff that could happen with intelligent systems watching
millions of security sensors so that you don't have to pay people to
monitor everything. So towns could install hundreds or thousands of
security cameras in the public places that would alert authorities whenever
something suspicious was happening (like spotting the young woman being
dragged off the path by the elementary school to be raped like just
happened around here this week), or reporting fires and accidents. There
are a lot of places that can justify paying for the cameras (which are
getting very cheap these days) but don't use them because they can't
justify paying for people to monitor them and after-the-fact recording of
crime doesn't help much if the prep knows he is being recorded and just
hides his identity. But if you could use automated AI systems to watch all
the video (and listen to the audio) - and pan and zoom the cameras as
needed to catch crimes in action and notify the police within seconds, then
there's not much physical crime people would be able to get away with. I
think if low cost AI systems could be developed to do that, it could become
wide spread in no time. And that would take a lot of intelligence in order
to correctly spot a crime in action and not get too many false positives
but it's the type of thing strong AI could do. These type of intelligent
surveillance systems I suspect might be possible in the next decade or two.

Which reminds me, I just recently read about an AI system that is used to
monitor swimming pools to spot people drowning which was just credited with
spotting a kid in trouble at a pool and saving their life. So some of this
is already starting to show up.

Here's a link about it...

http://news.com.com/Computer+system+said+to+help+stop+drowning/2100-1041_3-5558015.html

Smart surveillance technology is far more likely to be useful in the near
term than any type of smart car-driving technology.

The other spin-off of a technology smart enough to do surveillance is
search and rescue technology. Imaging having thousands of small bots smart
enough to search disaster areas to find all the people in trouble and
direct help to them. They don't have to be smart enough to fix them or
save them, just smart enough to do the search part. That's very likely to
be something we will see in the next decade. It's something that MIT swarm
project is already looking at.

But to do this stuff right, we really need strong learning AI systems here,
instead of all this hard-coded crap designed by humans. We need at least
dog-like intelligence to really do these search and rescue systems
correctly.

--
Curt Welch http://CurtWelch.Com/
curt@xxxxxxxx http://NewsReader.Com/
.

Quantcast