From: Peter Leaback <pete_leaback@cix.clink.co.uk>
Subject: Re: TECH: My standard is better than your standard!
Date: Mon, 29 Jun 92 22:42 GMT



An example of a VW was given that had two objects colliding. This is
ideally suited to next event time advance. The event is a change in
state of the world that occurs at a particular point in time. The 
event handler can predict the collision time and call the appropriate
object handlers and the object handlers can can pass a single message
to facilitate the correct action.

This style of implementation of a VW is insufficient for a general 
purpose system. We only know the actions of the user in fixed time
steps (eg the glove position is sampled), so the user could 
invalidate future events (eg get in the way of the colliding objects)

For this, and many other reasons I would suggest that fixed time step
time advance is more appropriate. An object handler would have to ask
the question "Has anything happened to me since my last call that 
effects my attributes ?". A way of answering this question is to 
assume that all objects are going to carry on in the same manner as 
they have been and see if this effects the object during the time
period in question. This becomes impractical for all but the simplest
of VW's. Another way of answering this question is to calculate the 
state of the system at an instant of time and assume that it is 
representative of the system over the whole of the time period in
question. One now enters the problems associated with of sampling.
 
If the time step is too small compared to the rate of change of an 
objects attributes, then one is wasting computing time. If the time 
step is too large, one gets an "undesirable" simulation of the VW 
(eg the two objects pass through each other). 

This method of time advance also means that the laws of the VW turn
a blind eye for a time period (eg the colliding objects are allowed
to intersect). The simulation would then have to fudge some 
corrective measures.

In a mature subject such as Digital Signal Processing, the effects 
of sampling and numerical resolution have been embraced by the 
subject. I feel that if we wish to formalise Virtual Worlds we must
address sampling and numerical resolution.


Pete Leaback.