Newsgroups: comp.robotics
Path: brunix!news.Brown.EDU!noc.near.net!howland.reston.ans.net!usc!elroy.jpl.nasa.gov!decwrl!csus.edu!netcom.com!nagle
From: nagle@netcom.com (John Nagle)
Subject: Re: Potential Field Methods
Message-ID: <nagleC8tv3E.CH@netcom.com>
Organization: NETCOM On-line Communication Services (408 241-9760 guest)
References: <1vnih1$qdl@elroy.jpl.nasa.gov> <1vpuhtINNaj5@ymir.cs.umass.edu> <1vq7ko$l8v@elroy.jpl.nasa.gov> <1vqg1jINNd9g@ymir.cs.umass.edu>
Date: Fri, 18 Jun 1993 17:25:59 GMT
Lines: 26

connolly@cs.umass.edu (Christopher Ian Connolly) writes:
>To be fair, I like their consideration of dynamic effects and
>stability.  They probably have a valid complaint that a lot of path
>planning work takes place in simulation.  Treating effectors as
>massless points only goes so far...

      Actually, the effective way to apply potential fields to path planning
seems not to be to apply the field to the vehicle position itself, treated
as a particle, but to apply it to a projected point of where the vehicle
is going to be if it continues its current direction.  The vehicle then 
chases that point.  Craig Reynolds devised this technique for his
"boids" bird-flocking simulation work, (it's in a SIGGRAPH paper) and
I've seen it re-invented twice since then, once for helicopter control
and again by John Connell for a mobile robot he built at IBM (not MIT;
this is newer work).  

      This concept needs a firmer theoretical foundation, but the ad-hoc
implementations thus far do provide smooth navigation in environments that
aren't dominated by obstacles.

      As for getting out of local minima, the dumb solution is to inject
noise, cranking up the noise level unti escape is achieved.  This is a 
form of simulated annealing.  The paths produced are awful, but it sort
of works.  Think of it as an implementation of "frustration" and "panic".

					John Nagle
