CS 490    SPECIAL PROJECTS IN COMPUTER SCIENCE:     Research Experience for Undergraduates

Fall 2007,  Instructor: Jeffrey Horn
 
 


  • HW1:   What's New in the Contreras Paper  (pass/fail)

    Students' answers:  (all on time!  all pass!)

    The paper didn't create original work, but what it found through a new 
approach was how do keep the same level of sensory response by using 
less hardware than prevoiusly used.
    		 
    I'm not really sure what's new about this article because I'm not really
up on the technological times, so to speak.  But I think that studying
phototaxis and it's relation to neural controllers is a pretty new idea. 
Especially when they're relating it to robots and how they could possibly
get robots to see.  This could actually be helpful with Fluffy's and my
project on making our mapping robot.  We shall see.  I have to look
further into this study.  But that's what I think the new idea is.
    		 
    By simplifying or restricting some neurons, there can be a positive 
result and roughly the same abilities in some cases.
    The use of direction sensors and a dynamic neural controller lead to the results that the direction sensors or a more well suited sensor for it's task can decrease the work load on the neural controller and less neurons in it were needed to perform the same task at a better proficiency than a panoramic sensor with more active neurons.

    New directional sensors, smaller neural controller was better than old panoramic sensor and bigger controller.
    In effect it sez a trophic hierarchy will enforce order in an evolving brain 
    By simplifying the amount of input, less processing is required to make
decisions.

    This paper is new not in that it evolves simple sensor configurations to get an agent to do something (like a simple Braitenberg machine), but in that it shows how certain sensor configurations that are similar to biological principles like ecological balance can eliminate the need for complex processing of stimuli thus making the controllers more efficient and easier to understand.
    This work showed advantages in using evolution for visual sensor controllers in search of deeper understanding with regard to the interaction between the agent and its environment.