Newsgroups: comp.sys.transputer,comp.realtime,comp.robotics
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!satisfied.elf.com!news.mathworks.com!uunet!news.cygnus.com!cygnus.com!rob
From: rob@cygnus.com (Rob Savoye)
Subject: Re: Need to control a robot-- Any suggestions?
Message-ID: <rob.793035472@cygnus.com>
Sender: news@cygnus.com
Nntp-Posting-Host: cygnus.com
Organization: Cygnus Support
References: <troybenj-1602952211460001@friley79.res.iastate.edu>
Date: Fri, 17 Feb 1995 15:37:52 GMT
Lines: 64
Xref: glinda.oz.cs.cmu.edu comp.sys.transputer:6775 comp.realtime:8719 comp.robotics:18131

troybenj@iastate.edu (Troy Benjegerdes) writes:

>for virtual reality. We are basically in the concept stage, and right now,
>we have an IBM-PC controlling a one-dimensional force feedback mechanism.

  Howdy, I used to design motion/vision controllers for robots. A PC 
actually works decently as a controller I've found, just get rid of that
DOS crap. I used Ctask once, a free light-weight thread kernel that runs
on DOS, and I used QNX too. (liked it!) If I did it again, I'd switch to
the VSTa free kernel as it's *very* similar to QNX (ANSI and POSIX), but it's
free. (ftp.cygnus.com:pub/embedded/vsta)

  PC's are ok if you find the right addon boards. Typically we went for boards
that had a lot of smarts on them to offload stuff from the PC. This is real
important for high-speed motion control. We once hit around 120 pick cycles
a minute, which was *fast*. 

>   I am looking for some kind of processing system to replace the PC. This
>system has to do several things: track the user's hand, calculate the

  This is just software. :-) You'd probably have to write alot of this
yourself, althought there are a few neural network based robot packages
out on the net. All this is the fun part.
  
>   This system should also have the capability to interface with a Silicon
>Graphics computer. We plan to use a SGI Reality Engine to do the graphics
>computation. We need some kind of system to control the robot that applies

  No problem. Just put a network card in the PC. Good motion control stuff
is made by Delta Tau (I think in San Diego). Much of what you  want to do can
be done with their motion control board. I also found the DataCube image
processing boards good for image processing at 60 frames a second.

>   The professor I am working with has heard of colleauges using
>transputers for this. I am also going to get some information from

  Sort of. Several folks are using transputers for the controller, but I'd
recommned against it mostly cause the development environments suck. The T-800
does context-switch incredibly fast though, but they can't handle large
programs very well. I once wrote an OS for transputers, and the tools are
really bad. I had to port C first cause the OCCAM is a waste of time.

>   One thing we need is at least 12 bit D-A and A-D accuracy (we think)
>and the ability to have one proccesor in the system communicate with the

  Lots of Digital A-D boards. Data Translation had good ones. One of our PC
based controllers had 380 I/O sensors on it. 

>   Another important consideration is cost. We don't really want to spend
>lots of money on something that may or may not work.
>   If anyone knew of a way to network SGI's to do this, that would work
>too. We have access to as many SGI's as we could use, but, from what I

  Don't use an SGI just cause you got one.... Our controller wasn't exactly
cheap, but it was 1/3 the cost of a proprietary (ADEPT) controller, and 15 
times as fast! I'd guess the basic hardware other than the motion board
is really cheap, but other than software it could be built for under $5k..

	- rob -
-- 
Cygnus Support        << KERNEL: Panic, core dumped >>             Headquarters
PO Box 1548                 Darkstar crashes,               1937 Landings Drive
Nederland, CO 80466    pouring its light into ashes,    Mountain View, CA 94043
+1 (303) 258-0506           reason tatters, ...               +1 (415) 903-1400
