Re: Randomness in Player/Stage simulation results

[email protected] wrote:

> Hi.  I've been running simulations (multiple  robots
> performing obstacle avoidance and exchanging messages) in
>  Player/Stage, and I notice that my code always gives
> slightly different  results each time I run it.
> 
> Where does this randomness come from?  Is there a random
> seed generated  somewhere deep within the code, or behind
> the scenes?  From a purely  theoretical POV, I'd expect to
> get identical results each time I run the  experiment, and
> that I would need to introduce some stochasticity into  my
> simulations in order to get a different result.
> 
> Does anyone have any guidance?

I guess that as long as we're not using hard real-time OSes, simply timing 
differences across runs will cause differences in the outcome. Different CPU 
loads will cause different thread scheduling; if message passing is via 
network there'll be also different delays in the message arrivals, and so 
on...

In player/stage this is to be expected, with stage running its simulation 
threads and player its driver threads, and the client another looping thread 
still.

In my experience, the opposite is true: to get exact repeatable results the 
easiest way is to use a single thread for everything to force a fixed 
sequencing, and even then take care of any I/O with anything out of your 
thread memory.

Another source of discrepancies could be floating precision... unless you're 
using fixed point.


------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first