Society of Robots - Robot Forum
Software => Software => Topic started by: macienne_12 on May 04, 2009, 11:25:09 PM
-
hi everyone, i'm new here with robotic
i have a problem with environment mapping with an ultrasonic censor
what mapping algorithm i should use?
i want to use dempster-shafer theory because the newer one, HIMM isn't able to use cause it's already used before
how to make mapping with dempster-shafer?
may anyone have a good tutorial?
thanks
-
How far are you with building the mapping software? It is important to have a system that can create the map from real readings taken from the sonar first. Once you have this you can "plug in" different navigation algorithms like the dempster-shafer algorythm or a simple random movement algorithm.
From what I know of dempster-schafer theory it could be used as a way of predicting either an unknown map segment or for guiding the robot to a goal within an unknown or unmapped area.
You are likely to require some other kind of navigation algorithm to start with. Part of Dempster-shafer is to determine whether enough data is known to accept a prediction or not, so starting a robot in a completely unmapped area the Dempster-shafer system should fail because there isnt enough data to provide a valid prediction.
Think like predictive text messaging on mobile/cell phones. Quite often the first letter typed doesnt offer any predictions its only after the 2nd or 3rd letter that predictions are offered, the predictions become more accurate if the system understands the context of the message, so before they work really really well at least a few whole words are generally needed to provide accurate predictions.
Look at how you are going to store the data and how data that is collected through the sonar can be compared and fed into a mapping system. If you make this a standalone closed loop then you should be able to plug in different algorithms like I mentioned above. Once you create a system like that then you can easily swap between different systems as and when necesarry, it will also be easier to modify the code involved if you only have to modify or upgrade a single plug in rather than the whole program.
Sorry if you have this all done already but the post is very generalised and doesnt offer any insight as to where you are up to.
edit:
-----------------------------------
Could you also let us know some specifics like whether your using a microcontroller(which one) or a computer and which language you are likely to be using.
Many people wont post if they dont know what capabilities you are working with
-
i haven't started yet with the software
the main problem is i don't really understand how to build a map with such method as himm, dst, etc.
in DST, does the belief function have to be calculate for all the grid from ultrasonic sensor?
in sonar sensor model, there are 3 region:
region I: grid probably occupied
region II: grid probably empty
region III: grid is unknown
in "Introduction to AI Robotics" by R.R.Murphy i found an example that the obstacle can be at region I or region II
so what does the function of region division?
does the range of each region is fixed? (ex: ultrasonic with 2.5 m range thus region I=0-1.5m; region II=1.5-2m; region III=2-2.5m)
i have a problem with this method of mapping...
hhmmmmmm,,,,are there any mapping method beside HIMM, Dempster-Shafer, Bayesian for ultrasonic sensor?
I'm looking for the simplest method for mapping the environment
-
So those methods are really all ways of trying to guess what is in unknown territories. They usually rely on trying to navigate to a destination.
There is nothing wrong with just using a random motion algorithm or a left wall following algorithm then building up a 2d array with real readings.
Once you get something simple like this working, you can then look at predictive mapping.
-
how to make mapping with dempster-shafer?
may anyone have a good tutorial?
http://vbn.aau.dk/fbspretrieve/14970824/Thesis_Alfredo230908.pdf (http://vbn.aau.dk/fbspretrieve/14970824/Thesis_Alfredo230908.pdf)
-
does the himm method can be use for dynamic environmental rmapping? so the obstacle can move, not stationary
anyone has a paper that explain the comparison time needed for building a map with such method as himm, bayesian, dempster-shafer?
thanks