Wednesday, September 28, 2005

evolving path integration on 2D agent

At the moment I am working on evolving a 2 dimensional khepera-style agent to perform photo taxis followed by path integration back to the place where it started from, controlled by a 2-5-2 node bilaterally symmetric continuous-time recurrent neural network (CTRNN). The simulation is a very fast version of a 2 dimensional world with simplified physics. It has however proven to transfer successfully to the real khepera robot.

There are a number of reasons why the path integration task is important for a dynamical systems approach to learning and memory but I will not go into it just yet. Today I've left running 6 sets of experiments. 5 evolutionary runs per set. 2000 generations per run. Results will be due in around 30 hours... this should give me time enough to work on my primary task the imprinting scenario which I will describe later on.

2 Comments:

At 3:28 pm, Blogger eduardo said...

as a follow up on the path integration: experiments ran until just now. the cluster was being used by other people - did not take that into account in my calculations... evolutionary runs look decent, some path integration evolved according to the fitness scores and it seems like non-symmetric agents did slightly better, as well as agents that could move both forwards and backwards.. more on this later. regarding the behaviours, well i've got to look at them live: i'm making a matlab movie out of some runs, and we've got thomas' 3D simulation as well... we'll see.

 
At 8:52 am, Blogger eduardo said...

on the technical side of things: looking into the evolutionary runs as well as behaviour several things came to mind about modifying/improving the simulation and task: (a) turning off the light once it is reached, (b) repair a bug in fitness function which causes it to temporarily shoot up, (c) try re-evolving and testing the best two agents (i.e. A1 and C1) in this new situation as well as testing them for longer periods of time and finally we should really try evolving agents for at least two 2 lights.

on the conceptual side of things: regarding the name of the task, path integration. what is path integration? are we talking about a mechanism or a behaviour? we say behaviour. let artficial evolution + situatedness + ctrnns suggest the posible mechanisms. moreso, surprise us!

this is all work in collaboration with thomas.

 

Post a Comment

<< Home