[ODE] Re: Virtual creatures (karl sims etc...)

Ian Sherratt (Shez) shez at texonet.co.uk
Tue May 14 07:54:01 2002

On Monday 13 May 2002 9:21 am, Nate W wrote:
> Both of those have drawbacks, though.  If you use the 'air time' detection
> method, you'll unfairly penalize creatures that try to leap from place to
> place.  If you use the 'spin' detection method, you might unfairly
> penalize creatures that try to roll from place to place.  There's still a
> of room for interesting behavior within those constraints though, so I
> wouldn't lose too much sleep over it.  If you find your creatures evolving
> right up to those limits, then it might be time to adjust the limits, or
> find different ways to guard against this 'cheat.' But, if you succeed in
> creating creatures that actually do evolve up to those limits, you've
> already had a huge success!
> But anyhow, other than this I'd be surprised if you ran into trouble using
> ODE's simulation to test your virtual creatures.  Do keep us posted on
> your progress, it sounds like cool stuff. :-)
I originally came across ODE whilst looking for physics engines to play with 
the same stuff, and have been interested in Sims's work for some time. 

Unfortunatly I got a real job before getting time to play with anything more 
than the graph representations of the creatures bodies and brains.  However I 
can tell you from past experience that GA/GP systems tend to be extrememly 
good at finding and exploiting inconsitencies found at the boundries of 
systems.  Karl Sims himself described his system as an excellent, if 
impractical tool for finding problems in physics simulators :-)

I remember one particularly cool example of this that was reported  when using 
genetic algorithm's to design simple circuits for FPGA's to run.   The best 
evolved solution involved 'components' that wern't even connected to the rest 
of the circuit, but removing them still broke the circuit....