Request for robotic simulation assistance

Request for robotic simulation assistance

[David Weller]

Real-time simulation of robotics requires contact detection to perform
dextrous "pick and place" operations.  In order to avoid visual image
pass-through of objects, this "detection of contact" must be considered
before the presented image is updated.

Ada and OOD concepts are being used to accomplish the actual
implementation of this task.  Polyhedrons are classified as either 
moving or static and compared to determine if contact has occurred.  Our
algorithm is two-stage: Possible collisions are determined by sphere-to-
sphere checking in the base coordinate frame, then edge-to-face checking 
reveals the point(s) of contact, if it exists.

Is there a more convenient form for collision detection between
objects in a graphical environment?  Additionally, our approach rests
heavily on sequential forms -- there is no tasking in our simple
model.  Is it prudent to build a graphical simulation of objects as
"tasks", particularly when real-time detection of collision (or 
collision avoidance) amongst all objects is important?  There is also
the outside chance that our simulation will work in a distributed 
environment -- will this impact design decisions?

(We're not asking you to do grunt work for us, we're just trying to
eliminate "deja vu" -- if somebody's already done this in Ada from an
OO mindset, we'd like to talk just to avoid mistakes and pitfalls)
 
All replies gratefully received, public or private.
Flames are not classified as replies -- please send to /dev/null

AdvTHANKSance.
D. Weller
----------------------------------------------------------------------
Mandatory .sig to eat bandwidth:
"Even barbarians like chocolate-chip cookies"
----------------------------------------------------------------------

Re: Request for robotic simulation assistance

[William J. Bouma]

In article <802@ajpo.sei.cmu.edu> wellerd@ajpo.sei.cmu.edu (David Weller) write
s:
>Is there a more convenient form for collision detection between
>objects in a graphical environment?  Additionally, our approach rests
>heavily on sequential forms -- there is no tasking in our simple
>model.  Is it prudent to build a graphical simulation of objects as
>"tasks", particularly when real-time detection of collision (or 
>collision avoidance) amongst all objects is important?  There is also
>the outside chance that our simulation will work in a distributed 
>environment -- will this impact design decisions?

  Convenient in what sense?  It sounds like you want someone to tell you
  there is a way to detect collision points without intersecting every
  pair of objects?  How could that be possible?  If objects are moving
  along preset paths you can sweep the time-space volumes and determine
  where these intersect to determine first time of collision.  Then you
  don't have to worry about the objects until your simulation progresses
  to that time.  But any interesting simulation will not have all objects
  moving along preset paths!

  Is your concern that the speed of this collision check is too slow?  If
  you use polyhedral objects, you can look at the Brep-Index methods of
  Dr. George Vanecek. (references if interested)  This yeilds a quick
  way of classifying the edges of one object against another.  You can
  analyze the classifications of all the edges to determine which entities
  of the two objects are in contact. 

  I am not sure what you are talking about when you say "tasking"?  What
  does "a graphical simulation of objects as 'tasks'" mean?

  Also, if you are doing this stuff in "real-time" you must have a bank of 
  super computers to run on!

  If you wish to distribute the work, obviously you can at least do the
  pairwise intersections in parallel.
-- 
Bill <bouma@cs.purdue.edu> 

    "The decision could easily be made by a computer." -- Dr. Strangelove