🚀 go-pugleaf

feng@srvr1 (Liqiang Feng, ) writes:

>I am working on the navigation problem for a tracked vehicle. 
  
      You need a heading reference.  A magnetic compass (flux-gate, 
presumably) or a heading gyro should do it.  You can often keep track
of distance travelled by dead-reckoning, but small errors in heading
have large effects after you've travelled any distance, so you need a
heading reference.

      					John Nagle

In article <1993Jan25.184635.11631@netcom.com>, nagle@netcom.com (John Nagle) writes:
> feng@srvr1 (Liqiang Feng, ) writes:
> 
>>I am working on the navigation problem for a tracked vehicle. 
>   
>       You need a heading reference.  A magnetic compass (flux-gate, 
> presumably) or a heading gyro should do it.  You can often keep track
> of distance travelled by dead-reckoning, but small errors in heading
> have large effects after you've travelled any distance, so you need a
> heading reference.
> 
>       					John Nagle


I am also working on this problem (in my head for now, implementation later).
I was thinking.  When the robot is activated, one could program the assumption
that straight ahead is zero.  Then, using vectors, one could compute direction
based on diameter of each wheel, voltage applied to each motor (velocity), and
change in acceleration.  This assumes a basic acceleration is known...of
course, the thing probably isnt going to go very fast at all, so one could
assume a constant acceleration.  Treat each drive wheel as a seperate vector,
then dead reckon via vector calculations.  Of course, it still falls prey to
small errors btw. model and real world, but if the robot gets 'lost', it could
just re-assume a zero and begin dead reckoning again.  This assumes the robot
has a target to reach.  This would imply a) the robot has an internal
representation of its environment (try that with fred's miniboard!), or there
is some external stimuli it is searching for.  If the latter, consider the
'lost' case one in which the target can no longer be sensed.  a vector can
be kept in memory keeping track of the target;  if contact is lost, assume the
last target vector as the new zero and compensate until target is regained,
then recalibrate the target vector.

Of course, I haven't done this, but it strikes me as similar to Rodney 
Brook's method in a few of his subsumption architecture robots...(from the
brief descriptions I've read and heard him give...)

'will think for food'

Mark C. Langston
langston@memstvx1.memst.edu (still inop)
graesserac@memstvx1.memst.edu (my prof's, I'm reachable via it)
fugue@mindvox.phantom.com (but I don't check this every day)

ttfn