Richard Kennaway jrk@CMP.UEA.AC.UK wrote£º
Wow! I’ve been working on a similar robot, also using PhysX for the
physics. I have one actuator for each of the three joints in each
leg. More details at Robotics and Perceptual Control Theory (which
I see I need to update).– Richard
Yes! Archy is an attractive robot. I did see it before, and it’s it who led me to PCT. How’s it now?
Rick, Thannks for your warmly reply! Glad meeting you and your wife in Guangzhou!
I should explain more about it but I didn’t. It’s a VC++7.1 project and still under heavy construction.
The best way to express how it works is demonstration. Attachment is a RAR pack which includes the video clip (WMV file, windows video format). I wrote a brief intro as well. Check the doc file.
What I mean in this question is: do we generate new controllers after we are born?
As you said, the most important thing to build is perceptual point of view from PCT perspective. I regard perception as trunk, and output linkages as leaves. The whole control system looks like a tree. Some parts of trees don’t have leaves. Does the number of leaves change? Does the trunk change? Or their architecture is fixed, only datas, connections and parameters on it change. If perception trunk is fixed, do controllers on it
change?
There’s different between hunger perception and perceptions in the action of fetching foods. I knew the system itself would never know the unit. What I want to learn is if this mechanism do something else. I mentioned in the last post, if this function has something to do with the connection between perceptions, connection may be rise to a high level, a single class, since lots of needed function related to it: re-organization, memory and so on. Connections between perception nodes are worthy to be studied more.
Best regards,
Thank you and your wife Linda’ warm regards!
Bo
Walking.rar (1.39 MB)
Robot Cricket.doc (165 KB)
[From Bill Powers (2006.09l.07.0741 MDT)]
Bo Wang (2006.09.06) –
Instead of commenting on the details of your cricket, I thought it might
be useful if I described my ideas of how this modeling project could be
organized. I think we are trying to do more than just make a cricket or a
cockroach walk so people can see it walking. That’s easy – you don’t
even need control systems to do that, just some clever links and rotating
cranks. For me, the problem is to explore how the neural control systems
might be organized, as if we were gods designing a bug. This should give
us insight about what the nervous system and muscles of a bug have to
accomplish, and that will be useful even if we don’t guess exactly
right.
Also, we must recognized that we are not the only ones working on this
problem: see
http://biorobots.cwru.edu/Projects/PRODUCTI.html. We have to ask what
PCT might contribute that others have not already done, and I think that
Richard Kennaway has done just right in focusing on the hierarchical
control aspects of the problem rather than on the details of leg dynamics
and kinematics. Ultimately the model must include everything, including
legs with mass and also including feedback from pressure/touch sensors
and limit sensors. However, it’s not possible to get to the final model
in one jump. Different people are working on different pieces of the
model, and nobody has the whole thing yet. Also, I think that PCT has a
lot to offer concerning the process of control itself – I came across
one review which said that while goal-seeking behavior involves feedback,
the reflexes are stimulus-driven.
Randall D. Beer wrote a book called “Intelligence as Adaptive
Behavior” which was actually about a cockroach model (if I remember
correctly). In it he proposed a “ring oscillator” which
generated the pattern of forward-and-back leg movements seen in real
hexapods. In HPCT, this might be a model of an output function at about
the “event” level (level 5). If treated in PCT terms, by adding
feedback control at this level, the ring oscillator might be a good
suggestion.
Bo, you ask how the details of the real system might have evolved. I
think that is a very hard question. We really can’t see how it evolved,
because it did not evolve in one step. It began, I would guess, with a
creature that could move cilia to bring food particles to itself in the
ocean. Each stage of evolution added some increased ability to control,
and at the same time altered the physical shape and physiological
equipment of organisms, so the next process of evolution could take
advantage of what was already there. The cockroach leg came from a
simpler leg, and that leg came from a tentacle, and the tentacle came
from cilia (possibly). What worked was retained, what worked less well
was changed. I don’t think we can realistically hope to model that
process because there are so many gaps in the historical record – and
anyway, if the same starting point could be recreated, there might be an
uncountable number of other end-points that are equally possible, that
would give the evolved life-forms just as good control over what happens
to them. We do need another data-point, from a life-bearing planet
orbiting a different sun.
I think we need to state our goals clearly. Just making a simulated model
walk is the least interesting part of the project – Walt Disney has
already done that just by drawing pictures. We want a model that will
generate walking and higher-level control processes, and that does it in
a way that takes real physics into account, and that uses circuits that
could exist in a real nervous system, and in the end, that resembles the
real system in all knowable ways. I think this is worth doing, because
when it’s done it will become much easier to take the next step, which is
to move up the evolutionary ladder. Stephen J. Gould said that the the
Tree of Life is really just a bush, but with PCT I think we can show that
it is a progression of life forms that have greater and greater ability
to defend themselves against selection pressures in more varied
environments – in other words, that can control the environment’s
effects on themselves in more ways and more kinds of ways.
Now I must get back to unpacking boxes and selecting the books I will
keep – about 1/3 of them can stay, the rest must go. Don’t worry, most
of those being discarded are murder mysteries, and the ones I keep will
be the ones that tell me the most about other countries and other
cultures. I never try to figure out who the murderer was. I just enjoy
the stories. What a long, slow, messy, confusing job it is, to move to a
new and smaller place. In another week, my apartment might be suitable
for public viewing. Right now, it’s not.
Best,
Bill P.
At 02:01 PM 9/5/2006 +0800, you wrote:
Hi, all
These days, I was practising PCT on my own simulated robot cricket whose
screenshots was attached to this mail. There’re two editions, and I’m
working with the white one now. The robot uses three servos to control 6
legs. Two legs in a group. I hope the robot could learn to walk towards
foods (yellow cubes) all by itself at last. And I have some questions
here:
- Do control system need new controllers?
Now that the objective of control is the perception, controllers may be
fixed to correspond perceptions. If the number of perception nodes is
fixed, controllers are already prepared as well, there’re only some
parameters to adjust. Though it sounds impossible, fixed controllers
can’t deal with the new challenge. However, in other part of human’s
body, some mechanisms do prepare for “everything”. Patterns of
geometry in the vision, antibodies (all discriminating proteins on
antigen are “considered”) and so on are such examples.
- In Bill’s reorganization experiment program, connections between
perceptions and output devices are in the same field. The perception of
joint angle is related to the output of muscle while the input from nose
can’t related to any output device directly. What about other
perceptions? Who decide which input could be connected to output
devices?Articles emphasizes only perceptions in the first level could be
connected to the output organs. Are all perceptions able to connect to
all ouputs? or just part to part
- Why only child, especially infants and babies, are able to rebuild
their neural systems when they encounter some physical problem? If an
adult is disabled for a long time, even if his organ could be cured, he
still lose his ability.It seems that re-organization of output device only happen when people
was young. Or like what cells do. An embryo has ability of turning into
any kinds of cells, while a mature cell lose this ability for good even
it has the same DNA with the embryo.However, study, another kind of organization, exist in man’s whole life.
Their principles may be quite different. Related to question 2, these two
sorts of organization may deal with different fields. Direct linkage and
indirect linkage
- What changed in our control system when we change my behaviour. When I
walk towards a bike and then ride on it, what happened in my
brain?Or behaviour is just linguistic. Human beings get used to analyse issues
with the usage of classification. May be in the brain, there’s no
distinct “borders” between actions that apears
different.
- How to convert the unit?
When I’m manipulating a joystick to control a simulated plane on the
computer screen, the translation from unit of the vision perception to
unit of the perception of my arm do exist. Consider it as a black box,
levels and transfer paths in it are ignored.Only the parameter “Ko” is enough? What if the controllers has
to send signals to tens of different controllers which have different
unit?To place the translation function on connections is a method. In this way
re-organization and memory (storing information and generating new
connections) could be integrated.Best regards,
Bo
No virus found in this incoming message.
Checked by AVG Free Edition.
Version: 7.1.394 / Virus Database: 268.11.7/435 - Release Date:
8/31/2006
Thanks for your concerning. I’m not mean to make trouble.
Here’s the .mov file compressed in .zip. I hope it works
Best,
Bo
Rick Marken marken@MINDREADINGS.COM дµÀ£º
Walking_Mov.zip (1.23 MB)
[From Rick Marken (2006.09.08.0930)]
.[From Richard Kennaway (2006.09.07.1637 BST)]
[From RIck Marken (2006.09.08.0840)]
Thanks, Richard. Now I have Media Player running but I can’t seem to
expand the wav file. Could you please send me a stuffed (or zipped)
version.Perhaps you don’t have a decompressor for RAR format? Try UnRarX from
http://www.unrarx.com/ (assuming you’re running Mac OS X).Thanks! I successfully downloaded UnRarX but couldn’t unstuff the
Walking.rar file. I think it’s becuase I have the oldest version of X.
As you
can tell, I’m pretty far behind the computer curve these days.Best
Rick
Richard S. Marken Consulting
marken@mindreadings.com
Home 310 474-0313
Cell 310 729-1400
Bill Powers powers_w@FRONTIER.NET wrote£º
Randall D. Beer wrote a book called “Intelligence as Adaptive Behavior” which was actually about a cockroach model (if I remember correctly). In it he proposed a “ring oscillator” which generated the pattern of forward-and-back leg movements seen in real hexapods. In HPCT, this might be a model of an output function at about the “event” level (level 5). If treated in PCT terms, by adding feedback control at this level, the ring oscillator might be a good suggestion.
I’m not sure whether cycle is basis of behaviour or not, especially on the forward-and-back leg movements. Repeated actions appears cycle, but I cannot say that’s the original purpose of control system. Cockroach model is much simpler than mammals, their mechanism of walking may be quite different from human’s. Is there
any “oscillator” in human body? Does cockroach nymphs learn to walk themselves?
I think we need to state our goals clearly. Just making a simulated model walk is the least interesting part of the project – Walt Disney has already done that just by drawing pictures. We want a model that will generate walking and higher-level control processes, and that does it in a way that takes real physics into account, and that uses circuits that could exist in a real nervous system, and in the end, that resembles the real system in all knowable ways. I think this is worth doing, because when it’s done it will become much easier to take the next step, which is to move up the evolutionary ladder. Stephen J. Gould said that the the Tree of Life is really just a bush, but with PCT I think we can show that it is a progression of life forms that have greater and greater
ability to defend themselves against selection pressures in more varied environments – in other words, that can control the environment’s effects on themselves in more ways and more kinds of ways.
Exactly, that’s what I want to see on this cricket, ability of learning, the generation of behaviours, not only walking around. Therefore, I made the model such simple(only 3 output) to increase the possibility of successful learning. In order to implement the progress of learning, architecture of memory has to be explored as well. Thousands of times “trial-error” may needed…And many other things to try
I wish everything goes right in your new home, though I can’t see it ; )
Best ,
Bo
You may disappointed, only angles on 6 shoulders are controlled in the video.
This is just a show of its gait, a part of introduction helping you understand its
mechanism(actually, you did, 3 motrs controlling 3 groups of legs are output
actuators of PCT controllers). The soul is under construction, since I’m trying to
make it learn walking all by itself and be different from a pure controller.
Best,
Bo
Rick Marken marken@MINDREADINGS.COM дµÀ£º
[From Rick Marken (2006.09.08.2110)]
On Friday, September 8, 2006, at 07:12 PM, Bo Wang wrote:
Thanks for your concerning. I’m not mean to make trouble.
Here’s the .mov file compressed in .zip. I hope it worksWorks like a charm. Thanks.
But I can’t tell what is being controlled. It looks like the front legs
move back and forth; the middle up and down and the rear one’s back a
forth in parallel with the front. Is the bug navigating to some goal
site? Or is it following some path on the grid? When it encounters the
cube (disturbance) it seems to just push it away; no avoidance.What is the bug sensing? What aspect of what is sensed is being
controlled? I think Kennaway’s bug senses and controls the position and
orientation of the body (it says so at his site anyway;-)) What does
the cricket control?Best
Rick
Richard S. Marken Consulting
marken@mindreadings.com
Home 310 474-0313
Cell 310 729-1400