Monday, October 03, 2005

circle-diamond categoriser with a reactive-controller

after implementing diamond-shaped objects into my circle-falling simulations (a la Beer) and intensive debugging i have gotten the simulation to work properly!!! also have managed to evolve so far reactive controllers as well as non-reactive ones for the task!!!

let's see how it goes.. there's much work to be done still. i shall remember to leave mass/inertia out of this one for simplicity. most likely the explanation of the evolved system is again one of historical interactions between the agent and its environment. the interesting question here is, what rules can be introduced into the evolutionary algorithm to select for simplest networks?

and would also be interesting to attempt evolution while vaying sizes of objects - mainly to find out (a) which is the simplest agent that can solve that task and (b) what is the dynamical systems explanation of the best-and-simplest evolved agent.

1 Comments:

At 11:16 am, Blogger eduardo said...

i've just started running the experiments described bellow, they should be finished by tomorrow morning...

Task: circle-diamond discrimination for fixed sizes

Variables of experiment:
- FF : whether the agent is controlled with internal state or without
- SYM: whether the agent's controller is symmetric or not
- INE: whether the simulated physics includes inertia for the body or not

About experiments:
8 sets of experiments (to include all combinations)
5 evolutionary runs per set of experiments (40 runs in total)
500 generations per run

The rest of the simulation uses the same methods as in Randy's Adaptive Behaviour's article:
- agent with a circular body diameter of 30
- agent's horizontal velocity proportional to the sum of opposing forces produced by two motors
- eye consists of seven rays distributed over a visual angle of pi/6
- magnitude of injected input when sensory ray intersects with an object in inversely proportional to the distance to the object along that ray [0,10]
- objects fall straight down
- fixed vertical velocity of -3
- initial horizontal offset from the agent in the range +-50
- circular objects with a diameter of 30
- diamond-shaped objects with sides of lenght 30
- agent controlled by CTRNN
- same fitness function
+ perhaps slightly different training examples (they use 24.. I use 21 per object)
+ when inertia/mass exists then mass of the agent is 2

 

Post a Comment

<< Home