From: harry.g.george@boeing.com   
      
   davejansen2@googlemail.com writes:   
      
   > Hello all,   
   >   
   > Does anyone know of a lab or a researcher that focus, or at least   
   > seriously tried, doing multiagent systems using evolutionary   
   > computation techniques (e.g., genetic algorithms, etc)?   
   >   
   > Will greatly appreciate any tip,   
   >   
   > Thanks,   
   > Dave   
   >   
      
   "Seriously tried" might be a stretch, but I did take a shot at this   
   for a MS CompSci project.  I took the RoboCup simulation system, wrote   
   a python script for general behavior, provided several variants of the   
   functions, and then did genetic programming to tune the selection of   
   functions used.   
      
   The main findings:   
      
   1. It took too long to run a single event.  I was using win/lose a   
      whole game.  Couldn't do enough events to allow evolution to   
      operate grdually, and thus had to tweak up my mutation and   
      crossover rates.   
      
   2. I didn't know enough about soccer (football) to write good   
      candidate behavior functions.   
      
   3. As far as I could tell, any approach which required thinking   
      (geneticically controlled or otherwise) was too slow to be   
      dumb-but-fast teams.   
      
      
   "Any tips": Don't make my mistakes :-).  Make sure you can test   
   success rapidly, so you can let evolution do its thing.   
      
   --   
   Harry George   
   PLM Engineering Architecture   
      
   [ comp.ai is moderated ... your article may take a while to appear. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)