Tried first the most vanilla version of the homeostatic adaptation task: 20 evolutionary runs with populations of 50 agents trying to do phototaxis to several lights. Nothing extra, simply as noisy as the one in Ezequiel’s paper – which happens to be noisier than anything else I had come across. In particular, the orientation of the sensors move about some, there is a lot of noise in the sensory input and motor output and the lights are quite far away. All of them evolved successfully to get to the light, but none out of the 20 best ones re-adapted after swapping the sensors around (in the first light analysis at least). Not too surprising.
Next, I will try selecting for the neurons to be non-saturated as well.