Oscar Gerelli wrote:
> we are trying to use the amcl localization algorithm in player-stage v.1.6.2
> but we got some problems.
> For the simulation tests we are using a pioneer robot within a map of our
> departement building, that is quite simple and regular.
> In our first test the localization error (the distance between real robot
> position in simulation enviromnent and hypotheses from localization proxy)
> seems to progressively increase during time.
> We notice that, in our scenario, particles spread over a huge area of the map
> instead of collapsing to the real position. We have no idea where is our fault
> in usind amcl localization.
> Could someone succefully in using amcl localization with simulator send us a
> simple test bed known to work?
Check the CPU load on your machine (e.g., with uptime, or top). If the
load is high, then amcl may not be getting all the cycles it needs to
process incoming observations. In that case, amcl will fall behind the
data, causing its pose estimates to get progressively worse.
Assuming the load is the problem, you can try:
- fewer robots
- lower resolution map (use the mapscale driver for this purpose)
- tweak amcl parameters to update less frequently and/or use fewer laser
beams from each scan
Alternatively, your problem could be caused by a mismatch between the
resolution (pixels per meter) used in the simulated world and that used
by amcl. Be sure you're giving the same numbers for both.
Brian P. Gerkey gerkey@...
Stanford AI Lab http://ai.stanford.edu/~gerkey