PCT robotics paper

[From Rupert Young (2016.09.21 22.40)]

A General Architecture for Robotics Systems: ** A
Perception-based Approach to Artificial Life**”

I am pleased to say that my paper has been accepted for publication

in the Artificial Life journal. It is basically applying
the PCT architecture to robotics, but also positioning perceptual
control as the missing ‘stuff’ of AI/AL (see attached).

It's a fairly long paper at 48 (book) pages (72 with refs and

appendices) with a fair bit of background of putting PCT into the
context of AI/AL, and a basic robotic experimental system.

**Artificial Life** is a major journal in the field so it will

be interesting to see the exposure and feedback it receives.
However, there’ll be a bit of a wait. I was going to annouce this
soon, when they sent out the contents for the Winter edition, but,
for some reason, it has now been bumped to the Summer edition next
year. So, I thought I’d let you know now, and I’ll send an update
nearer the time, along with pre-publication copies.

It's been a long road; by the time the paper is published it would

have been over three years since first submitted, but at least it
has now been accepted.

Regards,

Rupert

nature.pdf (112 KB)

[From Fred Nickols (2016.09.21.1752 ET)]

···

Yea! Congratulations, Rupert. Hard won and well deserved!

Fred Nickols, CPT

Writer & Consultant

DISTANCE CONSULTING LLC

“Assistance at a Distance”

View My Books on Amazon

Sent from my iPad

On Sep 21, 2016, at 5:41 PM, Rupert Young rupert@perceptualrobots.com wrote:

[From Rupert Young (2016.09.21 22.40)]

A General Architecture for Robotics Systems: ** A
Perception-based Approach to Artificial Life**”

I am pleased to say that my paper has been accepted for publication

in the Artificial Life journal. It is basically applying
the PCT architecture to robotics, but also positioning perceptual
control as the missing ‘stuff’ of AI/AL (see attached).

It's a fairly long paper at 48 (book) pages (72 with refs and

appendices) with a fair bit of background of putting PCT into the
context of AI/AL, and a basic robotic experimental system.

**Artificial Life** is a major journal in the field so it will

be interesting to see the exposure and feedback it receives.
However, there’ll be a bit of a wait. I was going to annouce this
soon, when they sent out the contents for the Winter edition, but,
for some reason, it has now been bumped to the Summer edition next
year. So, I thought I’d let you know now, and I’ll send an update
nearer the time, along with pre-publication copies.

It's been a long road; by the time the paper is published it would

have been over three years since first submitted, but at least it
has now been accepted.

Regards,

Rupert

<nature.pdf>

[From Rick Marken (2016.09.21.2150)]

···

Rupert Young (2016.09.21 22.40)–

A General Architecture for Robotics Systems: ** A
Perception-based Approach to Artificial Life**”

I am pleased to say that my paper has been accepted for publication

in the Artificial Life journal. It is basically applying
the PCT architecture to robotics, but also positioning perceptual
control as the missing ‘stuff’ of AI/AL (see attached).

RM: Congratulations Rupert!! How do we get a copy of the paper? The attached paper was by Rodney Brooks, who is not by any means a perceptual control theorist.

Best

Rick

It's a fairly long paper at 48 (book) pages (72 with refs and

appendices) with a fair bit of background of putting PCT into the
context of AI/AL, and a basic robotic experimental system.

**Artificial Life** is a major journal in the field so it will

be interesting to see the exposure and feedback it receives.
However, there’ll be a bit of a wait. I was going to annouce this
soon, when they sent out the contents for the Winter edition, but,
for some reason, it has now been bumped to the Summer edition next
year. So, I thought I’d let you know now, and I’ll send an update
nearer the time, along with pre-publication copies.

It's been a long road; by the time the paper is published it would

have been over three years since first submitted, but at least it
has now been accepted.

Regards,

Rupert


Richard S. Marken

“The childhood of the human race is far from over. We
have a long way to go before most people will understand that what they do for
others is just as important to their well-being as what they do for
themselves.” – William T. Powers

[From Rupert Young (2016.09.22 10.20)]

···

RM: Congratulations Rupert!! How do we get a copy of
the paper? The attached paper was by Rodney Brooks, who is
not by any means a perceptual control theorist.

[From Bruce Abbott (2016.09.22.0815 EDT)]

Congratulations, Rupert! I look forward to reading the paper.

Bruce

Rupert Young (2016.09.21 22.40) –

A General Architecture for Robotics Systems: A Perception-based Approach to Arti
ficial Life

I am pleased to say that my paper has been accepted for publication in the Arti
ficial Life
journal. It is basically applying the PCT architecture to robotics, but also positioning perceptual control as the missing ‘stuff’ of AI/AL (see attached).
It’s a fairly long paper at 48 (book) pages (72 with refs and appendices) with a fair bit of background of putting PCT into the context of AI/AL, and a basic robotic experimental system.
Arti
ficial Life
is a major journal in the field so it will be interesting to see the exposure and feedback it receives. However, there’ll be a bit of a wait. I was going to annouce this soon, when they sent out the contents for the Winter edition, but, for some reason, it has now been bumped to the Summer edition next year. So, I thought I’d let you know now, and I’ll send an update nearer the time, along with pre-publication copies.

It’s been a long road; by the time the paper is published it would have been over three years since first submitted, but at least it has now been accepted.

Regards,
Rupert

[Chad Green (2016.09.22.1153 EST)]

Fred, you’re also a participant on the EVALTALK listserv. A little over a week ago you posted this: “My favorite tagline is ‘Be sure you measure what you want.
Be sure you want what you measure.’â€?

Did you notice Bob Williams’ reply to my post on Sept. 7 (Re: Complexity) concerning the implications of the paradigm wars in the systems science community? Toward
the end he wrote:

“Because the management field failed to keep track of the important developments in the systems field in the 1970’s and 1980’s we now have a confusion of terminology.
So for instance, nobody I know in the systems field talks about ‘at the systems level’ when talking about very large management processes. That comes straight out of the management field, but creates enormous problems for evaluators who believe that ‘systems’
is solely about ‘big stuff’.â€?

Now take a look at your levels of HPCT example here:
http://www.nickols.us/LevelsofHPCT.pdf . Doesn’t level 11 also appear to reflect this outdated “at the systems levelâ€? construct to which Williams was referring? In other words, has PCT also failed to keep
track of the significant developments in the systems field?

Maturana’s notion of the self as a dynamic relation rather than a persistent object may be a good replacement candidate for HPCT’s Level 11. Here’s a relevant
passage from Maturana’s (1995) excellent paper Biology of Self-consciousness:

“As the self arises as an experience in the experience of self-consciousness, self-consciousness and self take place as dynamic relations in the flow of languaging,
and cannot be talked about without living them as experiences in the flow of language. The result of this situation is that all explanatory propositions that do not propose to treat the self as an entity (that can be ‘experienced’) seem off the mark. Strictly,
however, that is not the problem for the explanation, which as a generative mechanism only proposes a process that if it were to take place would give as a result the experience to be explained, and does not replace the explained experience as an experience.
But that the explanation should show that the self, self-consciousness and consciousness are but relational dynamics in the flow of our living as human beings, seems difficult to accept because we exist for ourselves as entities.”

Source: http://www.univie.ac.at/constructivism/archive/fulltexts/639.html

Best,

Chad

···

Chad T. Green, PMP

Research Office

Loudoun County Public Schools

21000 Education Court

Ashburn, VA 20148

Voice: 571-252-1486

Fax: 571-252-1575

“We are not what we know but what we are willing to learn.â€? - Mary Catherine Bateson

From: Fred Nickols [mailto:fred@nickols.us]
Sent: Wednesday, September 21, 2016 5:53 PM
To: csgnet@lists.illinois.edu
Subject: Re: PCT robotics paper

[From Fred Nickols (2016.09.21.1752 ET)]

Yea! Congratulations, Rupert. Hard won and well deserved!

Fred Nickols, CPT

Writer & Consultant


DISTANCE
CONSULTING LLC

“Assistance at a Distance”


View
My Books on Amazon

Sent from my iPad

On Sep 21, 2016, at 5:41 PM, Rupert Young rupert@perceptualrobots.com wrote:

[From Rupert Young (2016.09.21 22.40)]

A General Architecture for Robotics Systems: A Perception-based Approach to Arti
ficial Life

I am pleased to say that my paper has been accepted for publication in the
Arti
ficial Life
journal. It is basically applying the PCT architecture to robotics, but also positioning perceptual control as the missing ‘stuff’ of AI/AL (see attached).
It’s a fairly long paper at 48 (book) pages (72 with refs and appendices) with a fair bit of background of putting PCT into the context of AI/AL, and a basic robotic experimental system.
Arti
ficial Life
is a major journal in the field so it will be interesting to see the exposure and feedback it receives. However, there’ll be a bit of a wait. I was going to annouce this soon, when they sent out the contents for the Winter edition, but, for
some reason, it has now been bumped to the Summer edition next year. So, I thought I’d let you know now, and I’ll send an update nearer the time, along with pre-publication copies.

It’s been a long road; by the time the paper is published it would have been over three years since first submitted, but at least it has now been accepted.

Regards,

Rupert

<nature.pdf>

[Martin Taylor 2016.09.22.17,36]

[From Rupert Young (2016.09.21 22.40)]

A General Architecture for Robotics Systems: ** A
Perception-based Approach to Artificial Life**”

  I am pleased to say that my paper has been accepted for

publication in the Artificial Life journal. It is
basically applying the PCT architecture to robotics, but also
positioning perceptual control as the missing ‘stuff’ of AI/AL
(see attached).

Let me add my belated congratulations to the others. I think you are

right about “missing stuff”. I also note that Brooks talks about
perhaps going up to “five layers of artificial neurons rather than
today’s standard of three”. Powers has eleven, with the possibility
of layers within levels such as relationships of relationships (as
he discussed with me at CSG 93). The problem of setting all the
parameters for such complex neural networks is daunting when they
are seen as one-way machines for making sense of the sensory
environment, but not when each individual control “chunk” has its
parameters set by the need for good control. That’s a technical
detail, but an important one, in the same way as Powers’s discovery
of the e-coli process allowed him to reduce reorganization time from
the age of the Universe to continuous variation during a lifetime.

Again, congratulations.

Martin

[From Rupert Young (2016.09.23 21.00)]

(Martin Taylor 2016.09.22.17,36)

The problem of setting all the parameters for such complex neural networks is daunting when they are seen as one-way machines for making sense of the sensory environment, but not when each individual control "chunk" has its parameters set by the need for good control.

Yes, I think this is a very crucial point; that highlights a major difference between conventional wisdom and PCT. I see the problem, of trying to define a transfer function between input and output, in the real world, would only make sense if the system were a Laplacian Demon with entire knowledge of the state and dynamics of the universe, which is, of course, completely unfeasible.

For a long while I've been trying to come up with a simple, everyday analogy that would convey the difference between the adaptive nature of PCT and conventional open loop control with its hyper-sensitivity to parameters and the complexity of model-based control. Any ideas anyone?

Regards,
Rupert

[Roger K. Moore 2016.9.23.11.10 BST]

···

On 23 September 2016 at 21:07, Rupert Young rupert@perceptualrobots.com wrote:

[From Rupert Young (2016.09.23 21.00)]

For a long while I’ve been trying to come up with a simple, everyday analogy that would convey the difference between the adaptive nature of PCT and conventional open loop control with its hyper-sensitivity to parameters and the complexity of model-based control. Any ideas anyone?

You might be interested in the following which appeared as a (record length) footnote in a paper I published in 2007 …

“Heating an arbitrary room to a particular temperature requires the injection of just the right amount of heat based on the room’s size, the presence of other sources of heat and the means for heat loss. All this can be calculated analytically, but if any of the variables change, e.g. a window is opened or more people come into the room, then these disturbances would have to be sensed, their implications measured and the overall calculations repeated. Realising that such changes are unpredictable and happening all the time, and that the number of required sensors would get out of hand, the stochastic modeller decides instead to collect a database (in an attempt to capture the unexplained variability) with which to train a probabilistic system. The resulting device gives the right temperature 95% of the time (as long as the test conditions match the training conditions) but, in order to reduce the error rate even further, the only approach that is found to work is to collect more and more data. After many years of research, there is still a residual of variability that cannot be explained, and performance asymptotes. Through all this, it has been failed to notice that a simple thermostat would have quite adequately handled the infinity of possible conditions to a defined level of accuracy.”

Full paper here http://fulltext.study/preview/pdf/568942.pdf

Cheers

Roger

Regards,

Rupert

Ooops, sorry about the $$$ link - I can send a pdf if anyone’s interested.

···

On 23 September 2016 at 22:13, Prof. Roger K. Moore r.k.moore@sheffield.ac.uk wrote:

[Roger K. Moore 2016.9.23.11.10 BST]

On 23 September 2016 at 21:07, Rupert Young rupert@perceptualrobots.com wrote:

[From Rupert Young (2016.09.23 21.00)]

For a long while I’ve been trying to come up with a simple, everyday analogy that would convey the difference between the adaptive nature of PCT and conventional open loop control with its hyper-sensitivity to parameters and the complexity of model-based control. Any ideas anyone?

You might be interested in the following which appeared as a (record length) footnote in a paper I published in 2007 …

“Heating an arbitrary room to a particular temperature requires the injection of just the right amount of heat based on the room’s size, the presence of other sources of heat and the means for heat loss. All this can be calculated analytically, but if any of the variables change, e.g. a window is opened or more people come into the room, then these disturbances would have to be sensed, their implications measured and the overall calculations repeated. Realising that such changes are unpredictable and happening all the time, and that the number of required sensors would get out of hand, the stochastic modeller decides instead to collect a database (in an attempt to capture the unexplained variability) with which to train a probabilistic system. The resulting device gives the right temperature 95% of the time (as long as the test conditions match the training conditions) but, in order to reduce the error rate even further, the only approach that is found to work is to collect more and more data. After many years of research, there is still a residual of variability that cannot be explained, and performance asymptotes. Through all this, it has been failed to notice that a simple thermostat would have quite adequately handled the infinity of possible conditions to a defined level of accuracy.”

Full paper here http://fulltext.study/preview/pdf/568942.pdf

Cheers

Roger

Regards,

Rupert

very nice!

btw, use sci-hub.cc

it opens all paywalls!

···

On 23 September 2016 at 22:13, Prof. Roger K. Moore r.k.moore@sheffield.ac.uk wrote:

[Roger K. Moore 2016.9.23.11.10 BST]

On 23 September 2016 at 21:07, Rupert Young rupert@perceptualrobots.com wrote:

[From Rupert Young (2016.09.23 21.00)]

For a long while I’ve been trying to come up with a simple, everyday analogy that would convey the difference between the adaptive nature of PCT and conventional open loop control with its hyper-sensitivity to parameters and the complexity of model-based control. Any ideas anyone?

You might be interested in the following which appeared as a (record length) footnote in a paper I published in 2007 …

“Heating an arbitrary room to a particular temperature requires the injection of just the right amount of heat based on the room’s size, the presence of other sources of heat and the means for heat loss. All this can be calculated analytically, but if any of the variables change, e.g. a window is opened or more people come into the room, then these disturbances would have to be sensed, their implications measured and the overall calculations repeated. Realising that such changes are unpredictable and happening all the time, and that the number of required sensors would get out of hand, the stochastic modeller decides instead to collect a database (in an attempt to capture the unexplained variability) with which to train a probabilistic system. The resulting device gives the right temperature 95% of the time (as long as the test conditions match the training conditions) but, in order to reduce the error rate even further, the only approach that is found to work is to collect more and more data. After many years of research, there is still a residual of variability that cannot be explained, and performance asymptotes. Through all this, it has been failed to notice that a simple thermostat would have quite adequately handled the infinity of possible conditions to a defined level of accuracy.”

Full paper here http://fulltext.study/preview/pdf/568942.pdf

Cheers

Roger

Regards,

Rupert

I’m interested.

Fred Nickols

···

From: Prof. Roger K. Moore [mailto:r.k.moore@sheffield.ac.uk]
Sent: Friday, September 23, 2016 5:18 PM
To: csgnet@lists.illinois.edu
Subject: Re: PCT robotics paper

Ooops, sorry about the $$$ link - I can send a pdf if anyone’s interested.

On 23 September 2016 at 22:13, Prof. Roger K. Moore r.k.moore@sheffield.ac.uk wrote:

[Roger K. Moore 2016.9.23.11.10 BST]

On 23 September 2016 at 21:07, Rupert Young rupert@perceptualrobots.com wrote:

[From Rupert Young (2016.09.23 21.00)]

For a long while I’ve been trying to come up with a simple, everyday analogy that would convey the difference between the adaptive nature of PCT and conventional open loop control with its hyper-sensitivity to parameters and the complexity of model-based control. Any ideas anyone?

You might be interested in the following which appeared as a (record length) footnote in a paper I published in 2007 …

“Heating an arbitrary room to a particular temperature requires the injection of just the right amount of heat based on the room’s size, the presence of other sources of heat and the means for heat loss. All this can be calculated analytically, but if any of the variables change, e.g. a window is opened or more people come into the room, then these disturbances would have to be sensed, their implications measured and the overall calculations repeated. Realising that such changes are unpredictable and happening all the time, and that the number of required sensors would get out of hand, the stochastic modeller decides instead to collect a database (in an attempt to capture the unexplained variability) with which to train a probabilistic system. The resulting device gives the right temperature 95% of the time (as long as the test conditions match the training conditions) but, in order to reduce the error rate even further, the only approach that is found to work is to collect more and more data. After many years of research, there is still a residual of variability that cannot be explained, and performance asymptotes. Through all this, it has been failed to notice that a simple thermostat would have quite adequately handled the infinity of possible conditions to a defined level of accuracy.”

Full paper here http://fulltext.study/preview/pdf/568942.pdf

Cheers

Roger

Regards,
Rupert

[From MK (2016.09.24.1430 CET)]

Rupert Young (2016.09.23 21.00)]--

For a long while I've been trying to come up with a simple, everyday analogy
that would convey the difference between the adaptive nature of PCT and
conventional open loop control with its hyper-sensitivity to parameters and
the complexity of model-based control. Any ideas anyone?

Marionette vs unskilled marionettist at the outdoor theatre in windy weather.

M

[From Rupert Young (2016.09.28 15.50)]

(Roger K. Moore 2016.9.23.11.10 BST]

That's a very good way of describing it; and yes I'd like a copy of

the paper please.

It may be over-optimistic, but I'd like to come up with some snappy

one-liners to give an impression of the difference in difficulty and
complexity between the two approaches, such as,

... trying to whack a golf ball hundreds of yards into a little

hole, as opposed to moving the hole to where the ball falls.

Suggestions welcome.

Regards,

Rupert
···

You might be interested in the
following which appeared as a (record length) footnote in a
paper I published in 2007 …

          "Heating an arbitrary room to a particular temperature

requires the injection of just the right amount of heat
based on the room’s size, the presence of other sources of
heat and the means for heat loss. All this can be
calculated analytically, but if any of the variables
change, e.g. a window is opened or more people come into
the room, then these disturbances would have to be sensed,
their implications measured and the overall calculations
repeated. Realising that such changes are unpredictable
and happening all the time, and that the number of
required sensors would get out of hand, the stochastic
modeller decides instead to collect a database (in an
attempt to capture the unexplained variability) with which
to train a probabilistic system. The resulting device
gives the right temperature 95% of the time (as long as
the test conditions match the training conditions) but, in
order to reduce the error rate even further, the only
approach that is found to work is to collect more and more
data. After many years of research, there is still a
residual of variability that cannot be explained, and
performance asymptotes. Through all this, it has been
failed to notice that a simple thermostat would have quite
adequately handled the infinity of possible conditions to
a defined level of accuracy."

Full paper here http://fulltext.study/preview/pdf/568942.pdf

Content-Type: application/pdf;
  name="Moore - Spoken language processing Piecing together the puzzle.pdf"
Content-Disposition: attachment;
  filename="Moore - Spoken language processing Piecing together the puzzle.pdf"
X-Attachment-Id: f_itn2sbqr0

Moore - Spoken language processing Piecing together the puzzle1.pdf (411 KB)