simulation@uflorida.cis.ufl.edu (Moderator: Paul Fishwick) (09/21/88)
Volume: 5, Issue: 3, Wed Sep 21 12:30:26 EDT 1988 +----------------+ | TODAY'S TOPICS | +----------------+ (1) Real-Time O/S Kernels (2) ATM Node Simulator (3) State and Change...continued Moderator: Paul Fishwick, Univ. of Florida Send topical mail to: simulation@uflorida.cis.ufl.edu ----------------------------------------------------------------------------- Date: Mon, 19 Sep 88 09:37:23 MDT From: gatech!hao.ucar.edu!cmiller@bikini.cis.ufl.edu (Charlie Miller) To: simulation@bikini.cis.ufl.edu To: simulation@uflorida.cis.ufl.edu >From: cmiller@hao.ucar.edu Newsgroup: comp.simulation I was just curious if there are any other users of off-the-shelf real-time operating system kernels. Specifically real-time UNIX. We are using a product called VxWorks in the implementation of multi-tasking real-time pipelined video processors. I would like to correspond with other users of VxWorks and possibly share developed code and device drivers. =========================================================================== Believe it if you need it... -Charles E. Miller Software Engineer USPS Mail: High Altitude Observatory P.O. Box 3000 Boulder CO 80307 UUCP: {arizona,decvax,ihnp4}!hao!cmiller Internet: cmiller@hao.ucar.edu =========================================================================== ------------------------------ From: mcvax!ariadne!sstelios@uunet.UU.NET (Stelios Sartzetakis) Organization: FORTH - CSI, P.O.Box 1385, Heraklio, Crete, Greece 711 10 tel: +30(81)221171, 229368, fax: +30(81)229342, tlx: 262389 CCI Date: 20 Sep 88 11:43:51 GMT To: mcvax!cwi.nl!comp-simulation@uunet.UU.NET Subject: Submission for comp-simulation Responding-System: ariadne.UUCP Path: ariadne!sstelios From: sstelios@ariadne.UUCP (Stelios Sartzetakis) Newsgroups: comp.simulation,comp.protocols.misc Subject: ATM simulator wanted. Keywords: ISDN network ATD-multiplexing Date: 20 Sep 88 11:43:50 GMT Reply-To: sstelios@ariadne.UUCP (Stelios Sartzetakis) Organization: Institute of Computer Science, Crete I would like to ask if anyone out there has ever heard of an ATM (Asynchronous transfer mode) node simulator. -It's a specific packet oriented transfer mode using asynchronous time division multiplexing technique.- We are desperately seeking any kind of such a network node simulator. We will use it to process digital video data, for network management purposes. Reply to me directly, and if there is something of general interest, I'll summarize to the net. Thanx in advance, Stelios Sartzetakis Office: +30 81 229368, 221171,229302,229346 Fax : +30 81 229342...........(Preferred) Systems Analyst Telex : 262389 CCI GR...(if all else fails) Foundation of Res.&Tech Hellas UUCP :{mcvax,inria,unido}!ariadne!sstelios Institute of Computer Science P.O.Box 1385, Heraklio, Crete Greece 711 10 ----------------------------- Date: Mon, 19 Sep 88 17:11:40 EDT From: Dick Nance <NANCE@VTVM1.CC.VT.EDU> Subject: Time and State -- Recent Digest Issue To: COMP.SIMULATION Is the issue on non-linear behavior -- the article and the reference to Bernardo's book pertinent to discrete event enthusiasts or only to continuous types? The discussion on state and change has been an undercurrent in des modeling since the early days. Mike Overstreet's characterization of the three world views in terms of locality encapsulates the distinctions as well as can be done in few words: event scheuling: temporal locality activity scan: state locality process interaction: object locality The AI people have centered primarily on state locality, which our UK simulation colleagues have long held is the best for controlling complexity. My paper in Communications ACM, April 1981 dealt with the necessity for resolving much of the confusion surrounding the des modeling world. Suggestion: why not get some individual (or VERY small group) from each community (des and AI) to compose a list (10 or less) of the best works treating time and state relationships (state transitions, model indexing, etc.) to share with each other. These could prove valuable in getting a shared perspective, particularly on the major challenges of parallel processing for both communities or, more importantly, the intersection subcommunity. Dick Nance P.S. Keep my old address, at least for the time being. I am using this account to simplify the mailing. . QUIT z SMTP VTVM1 9/19/88 ' SMTP@VTVM1.CC.VT.ED NANCE@VTVM1 9/19/88 Undeliverable Mail ------------------------------ >From JMC@SAIL.Stanford.EDU Mon Sep 19 16:04:28 1988 Date: 19 Sep 88 1304 PDT From: John McCarthy <JMC@SAIL.Stanford.EDU> Subject: reply to message To: fishwick@FISH.CIS.UFL.EDU [In reply to message sent Mon, 19 Sep 88 11:13:16 EDT.] I would like to change the example slightly to make it clearer what capability we might want to give a robot. Suppose someone spills his glass on the table, and the robot is sitting at the table. It will have to do a triage on the papers on the table, deciding that some don't have to be rescued, others can't be rescued and some can be rescued if it acts quickly. My relevant experience is probably between 5 and 20 occasions. I don't disagree with your remarks on lack of a mathematical theory. The remarks on pattern recognition are vague, and I would be surprised if you knew how to make them precise. What patterns are meant? In particular which of the numerous aspects of an experience go into the pattern? There are two subcases. (1) You are permitted to design the pattern recognition framework knowing that saving papers from wetness is the problem. (2) Only your general annoyance avoidance mechanism is in question. Any liquid spilling specialization must come out of the general mechanism. Here are some concepts I consider relevant. 1. Epistemological adequacy. The general patterns or facts or other mechanisms concerning liquids must be capable of dealing with the information available to the robot at the time. We can suppose the robot has a TV camera and a microphone as its inputs. A lumped model can be provided if the work in forming it is done in advance, and you specify how the results of this preliminary physics is to be connected to the TV picture. 2. Elaboration tolerance. Suppose the particular table has trenches in it and that a squeegee is at hand. The idea of using the squeegee to sweep the liquid off the table or into a trench should be usable within half a second. I call it elaboration tolerance, because it involves the ability of the system to make use of new information. Another possible elaboration that might come to mind is someone saying to put an arm on the table to keep the liquid away from a particularly manuscript by Hubert Dreyfus. [In reply to message sent Mon, 19 Sep 88 11:13:16 EDT.] I think that using physics to build a lumped model is appropriate provided the result ends up in an epistemologically adequate form. As Genesereth and Nilsson said in their book, the problem of representing qualitative information about continuous actions and events is even more unsolved than the corresponding problem for discrete actions. Nevertheless, I suggest a contest to come up with the best description of what needs to be added to a robot to deal with the liquid spilling on the table and threatening the papers. I suggest that the entries be short enough to fit in paragraph or two and be readable by humans, but that the criterion for success be concreteness. I'll try but first I'd like to see a pattern or two. ------------------------------ Date: Wed, 21 Sep 88 12:18:47 EDT From: Paul Fishwick <fishwick> Subject: Reply to Prof. McCarthy To: comp.simulation The example that you provide concerning the pile of papers certainly sounds like a difficult problem; however, I would think that this problem should be made very specific so that one can get a handle on it. For instance, the problem might be stated as "sort through all papers and pick out only those papers that have some identifying characteristic." The problem must be stated unambiguously, otherwise, solutions will be too general and not specific enough for encapsulation into the robot. You are correct about my remarks on pattern recognition as being vague. They are vague simply because the problem of recognizing "shape from motion" in computer vision is a difficult one. However, this is what must be done if the robot is to succeed. Note that any theory on actions to take when a liquid is spilt presupposes solutions to the "moving liquid" pattern recognition problem. So when I discuss the pattern recognition problem, I am discussing a problem that must be addressed first before one deals with the manipulation of induced concepts at some higher level. To a great extent, then, I see theories of commensense reasoning as "jumping the gun." One must first solve the pattern recognition problem. I am certainly not an authority on computer vision and would be interested in hearing about problems associated with the recognition of moving liquids given frames obtained via a robots camera. I believe that most of the commonsense theories about the physical world are very interesting and belong in the domain of "philosophical inquiry." This is in no way meant as a slur -- philosophical inquiry has always been at the heart of knowledge and learning. I don't believe, on the other hand, that these theories have serious scientific merit since there is little or no attempt at validation. Thus, I am concerned when I hear "claims" of individuals having developed some new "qualitative calculus." Qualitative calculi can be obtained using lumped system concepts found in the systems and simulation area. Furthermore, the a qualitative model must have some clear advantage if it is to be proposed --- for instance a qualitative model derived through a homomorphic mapping which preserves behavior. A decrease in complexity is a good yardstick for measuring whether or not a qualitative model is useful. I am not sure how one would measure "epistemological adequacy." If someone can tell me precisely how to measure it then perhaps this criterion is a good one also. If a researcher can "study how a human would solve a problem" and derive an algorithm that is an improvement over standard methods then I think that this is great and reflects well on AI in general. An expert rule in MYCIN, for example, is important because it expresses knowledge otherwise unencoded (usually in a domain such as medicine where insufficient quantitative information is known). In physics or engineering; however, I remain unconvinced that commonsense theories have anything to offer yet. I often hear about "how easy it is is for humans to reason about basic actions." I find the premise "easy" to be most likely false. We don't know how we reason and we don't know how our eyes and brain can simultaneously perform simple calculations. The scientific approach would seem to be to concentrate on the computer vision problem first and only then build algorithms that "solve the problem" of spilt water. In summary, I would like to see someone take a problem like "detecting spilt water (or its trajectory)" from a more microcosmic viewpoint. Break the problem down into many subproblems each of which can be validated. Building a commonsense calculus or theory simply makes one wonder "What criteria can we use to test Theory ABC against Theory XYZ?" Let's crawl first, then walk. I like your idea of having a contest. Perhaps, through such a mechanism, we can get a better idea of the various subproblems that need to be addressed to answer questions that *appear* simple such as "How do I reason about water falling from the lectern?" -paul ------------------------------ +--------------------------+ | END OF SIMULATION DIGEST | +--------------------------+