Evolutionary Robotics

 
 

  1. Our approach:


    1. - On-line and on-body: individuals learn, i.e. evolve their individual controllers

    2. - “Real” mating: new morphologies derived from crossing over

    3. - Newborns shall learn...

    4.    -how to move

    5.    -to find energy sources

    6.    -to find mates

...before being released into the world


    1. - Open-ended evolution: robots just have to survive in a closed ecosystem

      1.    -but environment can be tailored to drive evolution towards desired behaviors

      2.    -energy is limited

      3.    -mates are not chosen at random, individuals decide wether to mate  depending on their relative skills

 

  1. Evolutionary Robotics -on the controllers side- deals with the automatic generation or adaptation of controllers for mobile robots. Traditionally, this has been done "off-line and off-body"


Modules:

-6 connection points (front, rear, left, right, top, bottom), 2DOF

-Head: 2 range sensors


Encoding: tree

  1. Phase 1: Learning by evolving controllers for modular robots

  1. Phase 2: Morphology crossover (in progress)