From: dstamp@watserv1.waterloo.edu (Dave Stampe-Psy+Eng)
Subject: Re: SCI: Latency in interplanetary transmission of telepresence
Date: Fri, 7 Feb 1992 02:07:23 GMT
Message-ID: <1992Feb7.020723.1260@watserv1.waterloo.edu>
Organization: University of Waterloo



tolman%asylum@hellgate.utah.edu (Kenneth Tolman) writes:

>to generate the environment.  This requires that the model matches nearly
>with the actual reality!  Thus the user could pick up a part of the space
>station and reattach it, and the commands for this motion would be sent
>out and later on the actual action is performed.  So the user operates in
>a "pre-world", always hoping that what they see themselves doing actually
>happens later.  When the model starts to deviate too significantly from
>the reality (not noticed until the signal has gone out and back) the model
>must be adjusted.
>
>  This has the severe limitation of only working when the reality is 
>already known and can be predicted under interaction.  An exploring robot
>that turned a corner and found a cliff (expecting flatness) would not update
>the users model until it had already fallen off.  However this technique has
>great potential and suggests research into modelling physical phenomena is
>needed to create accurately responding virtual environments.

What is needed is some more intelligence on the other end.  The telerobot
must be able to function on its own for extended periods of time, and 
recognize when it is in trouble.  I like the concept of the subsumption
archetecture for this: a hot topic with some robotic/AI poeple. 

In this case, the "upper levels" of the system are the human operator, plus 
some overall control programs on the telerobot.  In times of trouble, the
operator could "teach" the robot by controlling it directly.  The robot
would perhaps be able to learn the acton being performed, and use it in
future events.  (After all, this is the way you learned motor control at 
first:  human motor processing is very independant of "descriptive" learning
and learns best in the context of performing an action).  This builds
up a library of semi-autonomous motor programs.

Eventually, you might enter into a "shepherd-sheepdog" sort of relationship
with a number of telerobots.  You could command each where to go, and then
switch attention to another until it responds or is in trouble.  In times
of rreal trouble, you could "enter" the robot through telepresence and
use its manipulators and sensors directly (if time delay allowed).

--------------------------------------------------------------------------
| My life is Hardware,                    |                              | 
| my destiny is Software,                 |         Dave Stampe          |
| my CPU is Wetware...                    |                              | 
| Anybody got a SDB I can borrow?         | dstamp@watserv1.uwaterloo.ca |
__________________________________________________________________________