Genetic Programming Ant Behaviour Evolution

Particulars of the experiment: population size of 100, 40 generations, subtree crossover and tournament selection size 4. The behaviours chosen to visualise here were selected with the following criterion: the first improving behaviour (number of food collected) from each generation, where individuals tested in a random order.

Home

This ASCII movie shows the variations of behaviours evolved for the Artificial Ant problem using genetic programming. The Artificial Ant Problem requires genetic programming to find a strategy using motion and sensing commands that picks up food pellets on a 2D grid. The food pellets are marked with small dots in the visualisation on the left.

A behaviour is the sequence (ordered) of food collected. I do not show the actual ant moving on the grid. Instead, I am emphasising the different ways food is collected and how this varies over time.

Things to note from the visualisation:

  • Each ant starts from the upper-left grid position (a new behaviour),
  • The grid is torroidal (ants walk off the grid and appear on opposite size),
  • Food is sometimes collected in a natural-looking order,
  • And, sometimes it is not,
Return to Visualisations

Learn more about Genetic Programming

See my research concerning Behaviours, and other Publications

© 2004 Steven Gustafson