Advanced robotics

Martin, Got your request to join. Could email a brief intro for yourself to the group?
Bruce


Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

To further explore the complexities of the picture you are painting...
While the little OVT7620 camera is neat, it is nowhere close to what is needed for the sort of humanoid you are talking about. The first issue is this insistance that some robotics folk have on using the seriously flawed television video standards for machine vision.
Everything about television (particularly cheap television, CMOS Barbie-cam's, etc.) goes against good imagers for robotics: spatial resolution, temporal resolution, interlaced scanning, CFA focal-plane imagers, dynamic range, crippled color space. Television standards (NTSC/PAL and their digital variants) are seriously deficient in ALL of the above listed areas. Clearly, using such cameras will not result in superior next-generation vision of any sort, 2D or 3D.
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Martin Euredjian
  Click to see the full signature.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
On Sat, 30 Aug 2003 23:18:36 GMT, "Martin Euredjian"

Yeah, but try to process higher resolution images in real time! I would be exstatic to be able to do real time stereo matching on a 128 X 128 black and white image!
EdL
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
On Sun, 31 Aug 2003 03:07:30 GMT, "Martin Euredjian"

My point is that sometimes "Everything about television (particularly cheap television, CMOS Barbie-cam's, etc.) goes against good imagers for robotics" isn't necessarily true.
It's not enough to image using a camera; one must also process that video. High resolution can mean non-realtime processing.

I've put "inferior imaging technology" on a remote control vehicle and have been able to navigate just fine. That's because the human brain is able to process inferior images into meaningful information. The issue is that image processing technology is the more important thing than the imaging sensor in many cases. However, I agree that processing power is becoming more accessible.
EdL
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Remember the context of this thread: Bruce is talking about a "next generation" humanoid project. That ain't gonna happen with a set of little CMOS Barbie-cams.
One of the things I do for a living is design real-time image processing hardware. 2K x 1K 60fps, 10bit RGB (30 bits total) high-dynamic-range (logarithmically encoded) images can be manipulated in real time in many ways. Powerful FPGA's make it possible. Not a problem. Time and money.
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Martin Euredjian
  Click to see the full signature.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
On Sun, 31 Aug 2003 05:44:21 GMT, "Martin Euredjian"

There is nothing inherently wrong with CMOS Barbie-cams. There is nothing "Barbie-ish" about the technology residing in them. They provide great opportunities for cash-strapped robot experimeters. Showing the kind of disdain you do for them is not warranted with regard to hobbyists.

I too work on real-time image processing of the nature you describe.
However, remember the context of this thread: Bruce making wild claims about a revolutionary motor-replacing technology for which he had not provided any supporting information. As such, I thought he was a trolling confidence man and you seemed to be one of his sidekicks. It's easy to make false claims on the internet.
I've been doing newsgroups for over 10 years now and I have watched teams of trollers take over newsgroups several times. Their tactic is to set up an agenda and steer all conversations towards it. Their intent is always malicious and the results are the death of newsgroups by making real discussion impossible.
However, after examining his technology, I see that he has good ideas.
Sincerely, Ed LeBouthillier
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

You get me wrong my friend. I don't think there's anything wrong with CMOS/CCD "Barbie-cams". Especially not for the hobbyist. They are great, absolutely great. I have a dozen of them or so in different projects. Two of them are used in an eye-tracking goggle I designed for bio-feedback experiments. They are cheap, small, flexible. So, let's get that out of the way. However, you start talking advanced, next generation humanoid, advanced interaction with the environment and things are different. Using these imagers to try to develop higher-level solutions is a proverbial excercise in futility. A complete and utter waste of time. That's what I've been trying to say. Maybe this time I made it clear. I hope.

No, the context is the whole project. One for which absolutely no supporting data of any sort has been offered. Just wild claims. So, someone like me, who's done some pretty high-level motion control, image processing and related work, reads all of this and thinks "this guy is pulling our collective legs". So far, that's where my opinon sits. I think this whole thing is a joke. I might have some time this week to entertain looking at whatever might be in the Yahoo group. It bugs me though that it is OK to stirr things up in a public newsgroup and no information, absolutely no information at all is provided publicly to support these outlandish claims.

I think this is way out of line. How about we don't engage in name calling? I don't know this guy at all. I'm the one who voice strong skepticism about this whole thing being real. Read the thread.

Well, I've been doing newsgroups since before anyone knew the Internet existed. So what.
I realize that it is easy to loose sight that we might be talking to and with real people. I've taken the liberty to send you a private email with information on the company I own and run. I'm not this guy's sidekick, by far.

OK, how about some links? Or is everyone going to keep discussing this thing without one shread of evidence that any of it is real. I'll stand alone and say that what is being propose by Bruce is beyond the achievable today. And, that, based on the technology he is proposing the thing will not work and will provide any semblance of a useful humanoid. Humanoids are still way, way beyond what we can do in robotics.
Here's a simple test/task for Bruce and his team. Construct a wheeled robot (it doesn't have to be pretty) that can navigate any house and do just one thing: Use one or two arms (your choice) to make a cup of coffee and deliver it to the person who requested it. The environment, utencils and devices that this robot should use should be unaltered from those used by a human being. Make THAT happen and then worry about putting legs on the darn thing. This task alone could take a very capable team YEARS to resolve.
Enough said.
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Martin Euredjian
  Click to see the full signature.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Two
think
it
calling?
about
are
robot
a
darn
From reading the threads, it seems that all Bruce is asking is for people to come to the Yahoo group and see if they would be interested in helping, leaving the decision to them once they have the information. If that isn't backing up claims then I doubt that any picture or video would do better, since they could be all too easily faked ( I know, I've seen enough of them).
As a matter of fact, I find it very refreshing that someone is coming in with an open attitude, since typically these kinds of technical discussions are always too guarded due to the sensitive nature of the application and/or lack of (thought)/(common sense) in today's "high technology" (my personal feeling). It would seem that it could not get much more direct or open than asking if you want to help and be a part of it.
.02, for whatever that's worth (probably not much <g>)
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Are there any digital cameras that are useful? NTSC is nice for putting on a screen, but not too easy to use. I'd like to see something that reads out like a ram array.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
On Sun, 31 Aug 2003 19:11:55 -0500, "Dave VanHorn"

Although I have not tried it (and this is, therefore not an endorsement of the product), MVS has a cheap frame grabber that might be nice to use.
http://www.star.net/people/~mvs/TFGSPEC.TXT
Additionally, there is the CMUCam:
http://www-2.cs.cmu.edu/~cmucam/
It is available at a number of commercial outlets.
EdL
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

I have to apologize. As I am not in the consumer market I truly don't know what's available out there. Maybe someone can provide some ideas.
At the very least you want a progressively scanned imager (to avoid motion induced interlacing artifacts), as high a frame rate as your processing can handle, a digital interface and a full RGB color space (as opposed to YCbCr or YPbPr). Now, this will be expensive. If you are a hobbyist you might as well stick with some of the canned solutions as they are fun and cheap. If you can find an inexpensive progressively scanned camera go with that.
Maybe you can search Ebay for indusrial imaging cameras.
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Martin Euredjian
  Click to see the full signature.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
On Sun, 31 Aug 2003 20:56:19 GMT, "Martin Euredjian"

Clear enough. I think we're both sceptics of this guy's claims.

Yes, they're outlandish. I don't think he has the technology developed to a usable degree. Actually, what I've seen on his yahoo group is an electro-mechanical stepper motor/linear motor. I can imagine that it will suffer from severe friction, noise and be power-hungry.
If Bruce wants to develop the technology, he should do the math on the theoretical limits of his device. Then he will see where the problems lie.
As for the issue of "next generation," perhaps you should review what a generation might be. A generation might not be complete accomplishment of the goal, but merely a single advancement towards it. To me, a generation is about 4 or 5 years. Where will robotics be in 4 or 5 years? Therefore, CMOS sensors are legitimate technology for the NEXT generation (but probably not for 5 or 10 generations to come).

I don't think I was way out of line at all. Your statements about your opinions of him were the same as mine. My opinions were justified. If you read my statements, they were in the past-tense. I was not accusing you of being a troll-assistant, I said I thought (past tense) that you might be. I no longer think that.

The issue is experiences with watching newsgroups get taken over. My claim was that I thought this was one of those instances and therefore my point about observing years of newsgroups was support for my claim about being knowledgeable about such things.
Anyways, you and I are on the same side. We should set a policy together of de-escalating conflict so that we can work together to ensure that there isn't any trolling going on.

Good.
Look, perhaps the guy doesn't have the skills necessary for public discussion ( I think that's obvious ). However, you and I both agree that he made outlandish claims. I went to his yahoo group to see what information existed there. As such, I'm a little more satisfied that Bruce probably isn't malicious, but merely needs to learn some things.
Bruce needs to understand that there is a standard mode of public dialog which obligates those who make a claim to support their claim (most likely in the same arena that the claims were made). This is standard public communication and science reporting technique.
Bruce has been deficient in this aspect. He has made claims on this newsgroup but he has not provided adequate backup for his claim as he is obligated to do. People have openly challenged him and he has not responded appropriately.

Yes, the task is immensely difficult. I think that Mr Bruce has a lot to learn.
Cheers, Ed L
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Keep in mind the fact that we can navigate quite well in a video game. The display suffers the same losses as cameras of the day, and we can also use cameras for remote navigation of a waldo. So at least you have some basis for experimenting, even though it will not result in human style vision.
Cheers!
Chip Shults My robotics, space and CGI web page - http://home.cfl.rr.com/aichip
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

The
for
True, but this is only due to the superior cognitive abilities of humans. Sort of like driving in the fog. We can do it because we are able to process the seriously flawed information and get somethign out of it.
YOu have to keep in mind that a lot (years) of accumulated visual experience goes into being able to interpret flawed images. Something as simple as being able to tell that the shape behind a box isn't another object but rather the shadow cast by the sun requires pretty high levels of processing, particularly if done in real time, in motion and with all sorts of different objects (animals, plants, etc.)... Is that shape of a plane that's approaching me an object that's going to flatten me or just the shadow of an jet overhead?
I argue that you need lots more image data to even have a chance to derive real-world useful knowledge from a camera. I work with cameras costing in excess of $100K and, in some ways, they still lack what you need in advanced robotics (mostly frame rate and dynamic range).
The problem is that most people have never seen what a "real" camera looks like. Some of the better digital stil photography cameras can give you a sense. Dynamic range is not properly represented because you invariably endup in front of a 24bit (8 bit per channel) display. But, as a means for experimentation, someone could take their camcorder and a still camera and compare still images (you'd need video capture). The difference is huge.
So, if you were to put together a humanoid project, I would say that you'd need a team of a dozen accomplished experts JUST to develop vision technology that's conmesurate witht he stated goals. That will probably NOT cover the cognitive aspects of the problem. Just the hardware and some sort of processing engine.
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Martin Euredjian
  Click to see the full signature.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
snipped-for-privacy@earthlink.net (Ed LeBouthillier) writes:

Why do stereo matching? It's computationally very expensive. Most animals don't use it, they use the computationally much cheaper optic flow. And even animals which do use don't use it for high speed stuff.
-- Chris Malcolm snipped-for-privacy@infirmatics.ed.ac.uk +44 (0)131 651 3445 DoD #205 IPAB, Informatics, JCMB, King's Buildings, Edinburgh, EH9 3JZ, UK [http://www.dai.ed.ac.uk/homes/cam /]
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
On Thu, 25 Sep 2003 18:31:05 +0000 (UTC), snipped-for-privacy@holyrood.ed.ac.uk (Chris Malcolm) wrote:

Yeah, I'm not working on stereo vision now. Now I'm working on texture-based segmentation with a single camera. It seems more biologically plausible and should be computationally more efficient. Once I have a segmented image, even stereo should be easier to perform.
Cheers, Ed L
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
(Chris

Hi Ed,
Still playing with vision processing?
Stereo vision comes into its own when you have a random background of branches and leaves. It would be rather difficult to segment your random dot autostereograms.
Matching segments from two stereo images sort of works. The main problem I have is two segments in one image joining into a single segment in the other (or next) image or vice versa.
The other issue was segments that went beyond the left or right side of image. This made it tricky to calculate their centroids to determine depth.
For simple systems I think the practical solutions come from goal determined analysis. For example if your gaol is to detect and maybe even recognize people then movement is the easiest way to segment them from the background and determine their outline, particularly the head part which can act as a reference to the rest of the body.
As for texture based segmentation I find my images really don't show much texture unless the camera is very close to the surface or the textures are very large like patterns on wall paper.
One method that I found works is simple local thresholding the image into blob segments. I imagine these could be used as a framework for more intense analysis?
If you were interested I could send you an example? Maybe you could give some suggestions how to improve the code? If so let me know via the newsgroup. Any posts to this email address will go to spam land which I never download.
Cheers,
John Casey
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
wrote:

Yeah, not as much as I'd like to, but I'm still limping along.

That can be true. But a combination of texture and histogram can provide some useful info.

Yeah, I'd love to see what you're doing. Zip it up and send it.
Cheers, Art Ed LeBouthillier apendragn at earthlink.net
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
I am just concerned here, and perhaps it is a fault on my own part, but it appears( and I underscore appears) that every time a thread in this post gets too deep, it gets dropped. ( my news reader is less than adequate), so this may be my skewed version.
Your appearent reluctance to disclose your project in detail in this public forum leave me feeling uneasy. You have spoon fed us little pieces here, including the basic concept of your synth muscle. Could you please, for the sake of us all, post a digest of what you have revealed here, and perhaps everything else that you are willing to reveal.
This is my impression:
self contained "industrial" humanoid biped
Distributed processing, assumed to be serial, 2 mbps. Sub pocessor board needs to run at a bus speed of 20mhz. Sub processor board needs 3 x 10khz PWM to 3x 40 khz, Sub processor board needs power circuitry of 15 A, hundreds required ( Don't know if that is total aperage) ( Don't know what the bus voltage is) I think I read that you plan to water cool things?
3 phase liner motor, without rare earth magnets, just copper on copper, minimal air gap.
low resolution cmos images ( like 640x480)
I am not willing to join Yahoo groups because of the spam issues there, so I realize I am posting in a bit of a vacuum. I do wish to err on the side of caution when looking at your project, but your tactics make me feel uncomfortable. Generally, since this is a pulblic forum, we discuss things, well, publicly. Your responses seem to skirt different issues, leave incomplete answers, etc.
Part of what makes me feel uncomfortambe is that you use your own vernacular in your postings. SynthMuscle, FSys boards, AFAP (which I later figured out). If it is a 3 phase linear motor, call it that. There is a person here who posts regarding A.I. based on FORTH. I dare not invoke his name, because then this thread will be shot to H***, but he comes across as a complete nut job because he discusses his project in his terms, and assumes that everyone else can follow. He hides behind his own terminology, and won't discuss things in simple terms. Your project has a hint of that flavor.
Please understand, these are my own personal feelings, and I am only expressing them for your benefeit, not because a I am some bridge dweller. I am intrigued by your project, wouldn't mind helping, but I wanted to air my trepidation. Believe me, I have better to do than be a flamer or a troll. I have a literally 10 projects brewing, some with 4 year development times that are better served by my not being a news group Don Qixote, tilting lances with imaginary foes.
Again, I caution that this may be my skewed version.
Thanks
Mike
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Mike, here is your alternate access point:
http://groups.google.com/groups?&group=comp.robotics.misc
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Polytechforum.com is a website by engineers for engineers. It is not affiliated with any of manufacturers or vendors discussed here. All logos and trade names are the property of their respective owners.