DPRG
DPRG List  



DPRG: Diffrent terrains for navigation

Subject: DPRG: Diffrent terrains for navigation
From: Robert Posey robert.posey at worldnet.att.net
Date: Wed Aug 25 11:52:44 CDT 1999

- -----Original Message-----
>From: owner-dprglist at dprg.org [mailto:owner-dprglist at dprg.org]On Behalf
Of Bryan Andersen
Sent: Friday, August 20, 1999 3:26 PM
To: dprglist at dprg.org
Subject: Re: DPRG: Diffrent terrains for navigation


>Radar type sensors (IR Distance, Sonar, LiDAR, etc.) all are good
>for telling the distance and orientations to objects, but did the
>object move, or was it it's own motion that caused the distance
>change?  They are also prone to errors caused by the type of
>material being sensed.  They also require a much greater level of
>computation to use.

In the Book Sensors For Mobile Robots by H. R. Everett ISBN 1-56881-048-2

The problem comes down to 3 1/2 different cases.

  1. Robot has no prior knowledge of the the enviroment.

  2. Robot has an model of the enviroment.
    A. The enviroment is fixed or the Robot's model updated and thus the
       Robots Model is accurate, and the robot starts from a known position.

    B. The enviroment changes and/or the robots model is not accurate.

  3.  The Robot has a world model and there are special references to aid
    the robot in maintaining its own location and orientation in its world
    model.

If the robot doesn't have a world model it is a whole new ball game with
different problems.

If the robot has an accurate world model the main problem is locating the
robot in the model in terms of its X, Y and heading.  Assuming the robot
starts at a known location, the accuracy of its internal fix of its own
position will start to degrade the moment it leaves that position.  Using
Encoders, Compasses and other methods can slow the rate by which this
positional
error increases, but in time the error will reach a point where the world
model
is unusable, unless the robot can find a point of reference.  Managing the
error in robots internal knowledge of its location is one of the key points
of navigating a robot.

There are a lot of tricks to help
get fixes without modifing the enviroment.  One of those tricks is using
walls, doorways and intersections to relocate the robot in the model.
Doorways
are one of the best possible fixes, since a door's frame's center line is a
fixed point outside of Calfornia.  One important trick is to use the top
of the frame and not the sides due to the problem of opened door itself.

The idea is to use two optical sensors to detect when the door frame
is centered between them.  They are tuned so that they don't get a response
except when reflecting off the door frame.  Since they point up, this is
easy.

With 3 sensors as the Robot passes through the door first one sensor detects
the frame, the first one loses it and the second sees it, but not the 3rd
assuming the robot is lined up correctly.  Using two sets of these sensors
separated by a known distance, the robot can be aligned in the doorway by
turning until both sets have the door in the center position.  To the
missing
X position the robot can either use short range sonar, or simply use bump
by moving until it hits one side of the door frame.  Its best to use the
side the door does not connect to.  The bump method is also a cheap and
dirty, but fairly accurate method of getting at least one cordinate of
position for the robot from walls.  If the robot knows where it is in the
world at the begining, it can periodically bump walls to get fixes.  The
problem is that a wall can only get one cordinate at a time, and the quality
of your positional reference degrades quickly with the time between fixing
each cordinate.


If the world model is accurate, then the robot has a great advantage in that
it does not have to worry about what it uses for a positional fix of its
position.
All it has to do is find a way to get its position using its sensors.
Unfortantely,
outside of lights out manufacturing(no people) the enviroment will change.
This means the robot designer will have to worry about false fixes.  This
can be managed in many ways, two of which are.
  	1. Using references that are unlikely to change, or be blocked(i.e. the
top of
       a doorway).

      2. Keep track of the possible error in your position, and reject any
referencing
      attempt that returns a robot position that is outside the region
defined by
      this error.



If you can modify the enviroment.

Everett outlines several methods of providing cooperative fixes that are
easlier
than it would seem.  In a given room Sonar, RF of light sensors can be
used to find a mobile platform position within an inch.  With RF references
this can extend beyond one room.  There are several commercial systems, but
the sonar ones could be hacked with a little software expertise.  Another
way
is to have references that are passive reflectors of light or sound energy.
If you can make you references unique you can use 3 at a time to give you
a very percise location.  One important trick he mentioned was to place them
on the ceiling, since they are far less likely to be blocked by something.


I hope this helps,
Muddy

------------------------------

More information about the DPRG mailing list