DPRG
DPRG List  



DPRG: Diffrent terrains for navigation

Subject: DPRG: Diffrent terrains for navigation
From: Dave Everett deverett at idx.com.au
Date: Wed Aug 25 17:26:39 CDT 1999

At 15:24 25-08-99 -0500, Matt Minnis wrote:
>The type is Case #1
>
> > 1. Robot has no prior knowledge of the the enviroment.
> >If the robot doesn't have a world model it is a whole new ball game with
> >different problems.
>
>This is what I am working with trying to figure out.
>
Surprisingly this ca be the best way to go. I don't have much faith in
'world models'. Without prior knowledge your robot makes the least
assumptions about the environment.

>I want to build a map as I go, since it will be (eventually) taken to 
>contests, so it will have no good way to have an accurate map ahead of time.
>
>
>How can the diffrences between the compass and the encoders be 
>reconciled?  If I am right next to the filecabinet, or iron wood burning 
>stove, the compass will be off.  How do I know the compass is out of whack 
>and not to trust it?  How do I know that the encoders are correct since we 
>are on a ideal floor?
>
The trick is to use encoders and compass etc together rather than against
each other. All sensors should be used to build a series of landmarks that
are connected together. 

Think about what your sensors are good at detecting and use that rather
than try to make quantitative readings.

Odometry is probably the worst sensor on your robot so use that simply for
gross landmark discrimination and work your way up from there.

The best paper to get a feel for this type of thing is "Integration of
Representation Into  Goal-Driven Behaviour-Based Robots', IEEE Transations
on Robotics and Automation 8(3), 304-312 (1992).

In this paper a robot with 12 sonar, a digital compass and odoometry uses
simple behaviours to roam safely and builds a territory map based on sensor
readings. Later the user can command the robot to go to a location and the
robot will plan the path in realtime as it moves compensating for blocked
paths without explicit programming.

>How do you determine when a sensor is correct and when it is just wrong?
>
You do this with qualitative measurement not quantitative. Don't worry
about whether you are 15ft or 8ft from a target, worry about the quality of
the signal, limit your readings that way. If sonar is best below 2m then
stop listening after that.

Dave Everett.

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


More information about the DPRG mailing list