Tuesday 21 June 2011

Developer Journal 103


I left the simulation running overnight and the population did well. The population is growing slowly. I have a hunch I can increase the time robots have to wait before starting to reproduce and the time before being able to reproduce again.

elapsed real time: 10 seconds
elapsed sim time:  1.4 seconds
total real time:   17 hours 43 minutes 52 seconds
total sim time:    5 hours 5 minutes 50.3 seconds

Step: 741837 NumFood: 8 NumFoodEver: 1247 NumCrit: 85 NumCritEver: 210
numSick: 0
numChild: 0
0 to 10000: 10
10000 to 20000: 4
20000 to 30000: 1
...
skip a few
...
730000 to 740000: 0
740000 to 750000: 5

110621-1404 (26 MB) is the latest save file.

Hmm. When robots move their view ports wobble too much because they have a cylinder shape. I originally moved from a cube shape to a cylinder shape because the robots were getting stuck on walls. I'm going to go back to a cube shape for a while. There's definitely less view port wobbling but I don't know whether the population can figure out how to not get stuck on the walls.

Whoa. I added a wait time for how often the brain gets updated. Normally there is one brain update per frame update. That's quite intensive. So the base line is one brain update per 1 / 30 = 0.03 seconds. I tried one update per 0.5 seconds. The difference was as expected, huge. 10 updates per second is pretty good as well. Sticking with this for now.

Hmm. Switching to a block shape causes more robots to be stuck on the walls. Changing from 30 brain updates per second to 10 makes the robot behaviors more intentional. The processing power saved from less brain updates per second is great and things seem okay for now.

I'm looking forward to the following next actions

+ only updating the view ports when I update the corresponding neural network
+ stagger neural network updates
+ increase robot vision resolution
+ experiment with more humanoid body shapes



RELATED

No comments:

Post a Comment