Newsgroups: comp.robotics
Path: brunix!doorknob!jak
From: jak@cs.brown.edu (Jak Kirman)
Subject: De[a]d reckoning and robot navigation
In-Reply-To: hollombe@polymath.tti.com's message of 9 Apr 92 19:07:57 GMT
Message-ID: <JAK.92Apr11220313@bimini.cs.brown.edu>
Sender: news@cs.brown.edu
Reply-To: jak@cs.brown.edu
Organization: Dept. of Computer Science, Brown University
References: <hNk+rwB@engin.umich.edu> <34409@ttidca.TTI.COM>
Date: Sun, 12 Apr 1992 03:03:13 GMT
Lines: 55

In article <34409@ttidca.TTI.COM> hollombe@polymath.tti.com (The Polymath) writes:
>   Just a nit:  The correct spelling is "ded reckoning". "Ded" is short for
>   "deductive." The term usually refers to navigation by reference to known
>   landmarks, so, yes, it's still a feasible option, but I don't think that
>   was the question you were asking.

Well, I love nits, and this one has aroused my curiosity.  My OED gives:

   Dead reckoning. Naut. [DEAD a. V.]  The estimation of a ship's
   position from the distance run by the log and the courses steered by
   the compass, with corrections for current, leeway, etc., but without
   astronomical observations.

There follow a number of quotations using the term, dating back to 1613.
[DEAD a. V.] specifies that DEAD is being used as an adjective in sense
V, which is:

   Unrelieved, unbroken; absolute; complete; utmost.

What sources do you have that give "ded reckoning"?

According to the definition above, navigation using landmarks is not
dead reckoning.  I think the common usage agrees with the definition
above. 

Finally, to give my opinion on the original question, if you only use it
for short periods of time, on surfaces that are not too slippery, and
your robot's wheels don't get out of alignment much, and the surfaces
are level, and you don't need measurements that are too accurate, it
works wonderfully.

We have a couple of Real World Interface 12 inch bases.  We use them in
an office environment -- carpets on which the wheels don't slip too
much, and fairly level floors.  The principal problem is alignment: the
robot has three wheels, and over time they get further and further from
being parallel.  So when the robot thinks it is going straight, it is in
fact describing a broad arc.  Such angular errors quickly produce large
positional inaccuracies.  There is also something of a problem with
uneven floors: if there is a bump positioned such that one side of the
robot goes over it but the other one does not, then the robot ends up
turned through a small amount, without realizing it.  There are a few
places where this has caused problems for us.

Basically, we rely on dead reckoning only for the coarsest of
measurements: we use a simple PID controller to drive the robot down the
middle of a corridor to a junction, and dead reckoning to see whether we
somehow missed a junction because a door was closed or people were
blocking off a passage, etc.  When it is a matter of knowing whether you
have traveled 4 meters since the last junction or 5, dead reckoning is
fine for the most part.

                         Jak Kirman                        jak@cs.brown.edu
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Any clod can have the facts, but having opinions is an art.
                                   --Charles McCabe, San Francisco Chronicle
