From dblank at brynmawr.edu  Thu Apr  3 11:26:18 2008
From: dblank at brynmawr.edu (Douglas S. Blank)
Date: Thu Apr  3 11:26:45 2008
Subject: [Pyro-users] Numeric memory leak work-around
Message-ID: <47F4F71A.1030302@brynmawr.edu>

For those that use the neural network package (conx.py) in Pyro, you 
might want to update from the latest Pyrobot 5.0.0 (either from tar 
file, or CVS). We recently found a memory leak in modern Numeric 
versions and have provided a work-around in our code.

(The problem was a memory leak in the Numeric function tolist(). We have 
changed all calls from .tolist() to .copy() and that seems to have 
eliminated the memory leak).

-Doug

updates available from:
http://pyrorobotics.org/download/
http://pyrorobotics.org/?page=PyroDeveloperCVS
From D.Zoltok at sussex.ac.uk  Mon Apr 14 11:14:39 2008
From: D.Zoltok at sussex.ac.uk (Dave Zoltok)
Date: Mon Apr 14 11:14:56 2008
Subject: [Pyro-users] Maintaining trails
Message-ID: <301061A7-1DFA-4BE8-9087-41958BD74B24@sussex.ac.uk>

I'm currently running experiments with evolved robot controllers and  
I'm trying to see the robot's trails during multiple attempts at  
navigating towards a light source. At the moment this involves  
randomizing the robot's initial state, placing it in the environment  
and letting it run, and repeating the process several times with  
different randomizations. The problem is that whenever I reset the  
position of the robot, the trail vanishes, and the movement from the  
final position of one trial to the initial position of the next trial  
becomes the first segment of a new trial.

Is there a way to have trails persist in the environment between  
resets? I've tried to piece together how the code is running, and it  
seems like I would need to store every generated trail separately, so  
they can be drawn independent of the robot's position. I have no  
problem doing that, but is there a better way?

Dave
From dblank at brynmawr.edu  Mon Apr 14 13:21:43 2008
From: dblank at brynmawr.edu (Douglas S. Blank)
Date: Mon Apr 14 13:21:54 2008
Subject: [Pyro-users] Maintaining trails
In-Reply-To: <301061A7-1DFA-4BE8-9087-41958BD74B24@sussex.ac.uk>
References: <301061A7-1DFA-4BE8-9087-41958BD74B24@sussex.ac.uk>
Message-ID: <480392A7.9080602@brynmawr.edu>

Dave Zoltok wrote:
> I'm currently running experiments with evolved robot controllers and I'm 
> trying to see the robot's trails during multiple attempts at navigating 
> towards a light source. At the moment this involves randomizing the 
> robot's initial state, placing it in the environment and letting it run, 
> and repeating the process several times with different randomizations. 
> The problem is that whenever I reset the position of the robot, the 
> trail vanishes, and the movement from the final position of one trial to 
> the initial position of the next trial becomes the first segment of a 
> new trial.
> 
> Is there a way to have trails persist in the environment between resets? 
> I've tried to piece together how the code is running, and it seems like 
> I would need to store every generated trail separately, so they can be 
> drawn independent of the robot's position. I have no problem doing that, 
> but is there a better way?

Dave, I wrote that and I have no idea. I'd have to go back and see how I 
did that, and what resets it.

If you come up with a better-defined interface for handling the trails, 
please submit the code. Thanks!

-Doug

> Dave