ARIA-Users

Date Index Thread Index

Re: [Aria-users] arnl



Matthew Johnson wrote:
> We are using Aria2.5.1 and SonArnl1.5.1 both in Java.  On the robots
> (running linux) It is working fine, but takes a long time to find the path
> (~10 seconds).  Is this normal?  What determines how long it takes to find
> the path?

Well, the released version of ARNL does a complete search before it 
finds the path, thats why its taking so long.  This means that it 
expands the path to fill everywhere accessible.  If you put forbidden
lines around where the robot can't go that can speed it up a lot.  I'd 
need to see the map to know more about it.

We've done some changes recently so that ARNL will stop searching when 
it finds a path, which saves time in situations like that.

Other than that, how long it takes to find a path is determined by 
planres, note that the its a squared relationship, so if you go from 100 
to 50 it takes 4 times longer, or 100 to 200 takes 1/4 as much time.

> Also, we run the same thing in simulation on a windows laptop.  It seems to
> load the map fine, but never finds a path.  We get:
> 
>  
> 
> Loaded map file 'BravoPier.map'
> 
> Path Planning to pose (2000 1444)
> 
> Bad grid lengths
> 
> Bad grid lengths
> 
> Bad from and/or to points for plan
> 
> PlanAndSetupAction: No path from Current      0     0     0 to Goal  2000
> 1444     0 
> 
> Goal Failed:  2000  1444
> 
>  
> 
> This map works fine on the actual robots, so we are not sure what is gong on
> here.  Any suggestions?

The bad grid lenths means that the path planning really didn't get set 
up.  I think that it's probably running out of memory... either because 
its a huge map and not enough memory, or because PlanRes is really low 
so it takes up a lot more memory.

Matt LaFary
MobileRobots Inc