Newsgroups: comp.robotics
Path: brunix!news.Brown.EDU!noc.near.net!howland.reston.ans.net!usenet.ins.cwru.edu!agate!headwall.Stanford.EDU!CSD-NewsHost.Stanford.EDU!CS.Stanford.EDU!mark
From: mark@CS.Stanford.EDU (Mark Yim)
Subject: Re: Potential Field Methods
Message-ID: <1993Jun21.200105.3993@CSD-NewsHost.Stanford.EDU>
Keywords: potential fields, DP, spurious minima, local minima, global minima
Sender: mark@CS.Stanford.EDU (Mark Hosang Yim)
Organization: Robotics Department, Stanford University, Ca. USA
References: <1vtedj$55l@elroy.jpl.nasa.gov> <1vtlk2INNsqh@ymir.cs.umass.edu> <200160$mkk@elroy.jpl.nasa.gov> <202gb9INNkkb@ymir.cs.umass.edu>
Date: Mon, 21 Jun 1993 20:01:05 GMT
Lines: 31

In article <202gb9INNkkb@ymir.cs.umass.edu>, connolly@cs.umass.edu (Christopher Ian Connolly) writes:
|> In article <200160$mkk@elroy.jpl.nasa.gov> gat@robotics.jpl.nasa.gov (Erann Gat) writes:
|> >I gather that it is essentially equivalent to a spreading-activation method
|> >of computing the shortest path distance to every point in a region in parallel.

For the Barraquand/Latombe method. They use a wave front propagation method (NF1).

|> Actually, I was curious about this awhile back & coded up a dumb
|> version of the algorithm (just repeated sweeps) and found that it's
|> not that bad -- a 60x60 grid (which is about a 6 inch resolution on
|> our lab floor) takes about a half second on a DECstation 5000.  At
|> this speed, I/O to the robot + movement tend to be the bottlenecks.

Woah,  60x60 grid should probably take less than a millisecond on a Dec 5000 with 
Barraquand/Latombe method, probably less than a second on an 8bit micro.*

|> > second, it requires recomputation from scratch when the environment changes.
|> 
|> This is one of our pet peeves too -- it doesn't seem to behave
|> gracefully with the addition or deletion of obstacles and goals.  

Actually some people in our group are doing path planning with
environmental changes (moving obstacles around, moving goals)
Because it's so fast you can replan from scratch very quickly.

* DISCLAIMER:
I'm not personally involved in any of the above path planning stuff...
(so, my times could be off by +/- one order of magnitude :-)

mark
mark@killdeer.stanford.edu
