The last post was 48 days ago on Tuesday 2nd August 2011, Developer Journal 119. The past few weeks have been about getting my act together and getting over being sick. Coding and experimenting came to an almost standstill. Now, some basic experiments have to be done before work can continue on connecting the neural network and the robot arms and legs.
A few ideas came to me while at the 2011 Australian Singularity Summit, about how to more easily modify the neural network to allow additional inputs and outputs on the fly. Part of me is eager to implement those ideas and another part of me is hesitant to face the complexity.
The way child neural networks combine from parent neural networks still doesn't seem right to me but I don't know what to do about it.
Grr, loading up a file with 247 robots takes 8 minutes, 17 seconds. A new computer would be great about now with a faster processor, more memory, and a solid state drive.
It's also been about 48 days since I've run any experiments. I ran a experiment with the population that's been ongoing for a while yesterday. I expected that the population should survive but it died and I'm worried because I'm not sure what's changed.
The save file I continued from had 56 robots so I went back to an earlier save file with 247 robots, culled the population to 128 robots, and continued the simulation.
I ran a few experiments over the past few days where I had changed a few things and the populations died out so I want to make sure that the original conditions still resulted in a population that can persist.
Grr. A few robots are getting stuck on top of the walls! They can't move off the walls because I disabled their ability to move when they're above the ground by a little bit. This was to prevent robots from flying off the ground when they face away from the ground. I increased the wall height to 10 metres and changed the code to reset robot positions if robots get higher than 10 metres which they shouldn't be able to.
RELATED
No comments:
Post a Comment