Saturday, August 7, 2010

Evolving a self-organizing soccer team

This video shows the evolution of coordinated behavior of simulated robot soccer players. In the simulation, each soccer player is controlled by a neural network. The neural networks are evolved using an evolutionary algorithm, so generation after generation the strategy improves.
After a few hundred generations, the players of a team adopt a useful behavior. The used approach did not include a trainer telling them how to play or specifying predefined roles for the players such as being a defender, midfielder or striker. Still, during a game, different behavior of the players emerges. Thus, similar to biological systems, the entities take up different roles in a self-organizing way. Since the agents are not predefined, such systems have a high robustness against failure of come of the entities.



16 comments:

  1. what happens next? did you manage to tweak the algorithms so better overall strategies are reached? specialize units as defense, midfield, etc?

    ReplyDelete
  2. Already more interesting than real football.

    ReplyDelete
  3. To say this is similar to any biological system would mean that the algorithm would evolve the hardware and software systems.

    Not just the algorithm!

    ReplyDelete
  4. That's right. As a further step we are working on co-evolving hardware and software for robotic systems by integrating Lego Mindstorms models into a simulation environment with our FREVO tool. The modularity of Lego allows an easy change of physical properties.

    ReplyDelete
  5. I don't understand how they could learn football at all, without some incentive to play with the ball, put the ball in the goal, etc. If there is no training or incentive for them to play, how did these systems evolve to learn football? Why didn't they learn some other game?

    ReplyDelete
  6. The incentive is encoded in the fitness function. Basically, the fitness is the higher the more goals they score over their opponent. In order to make the fitness function smooth enough for evolution, some subgoals like covering the field, having ball, kicking the ball and getting the ball near the opponents goal were defined. This is some help, but still they had to figure out how to play the game.

    ReplyDelete
  7. Are you continuing with the experiment and introducing new rules, like the offside rule, to see how the players handle changes tot he system?

    ReplyDelete
  8. Aren't they doing off-side passes there?

    ReplyDelete
  9. Is there download for this model for FREVO available?

    ReplyDelete
  10. We are currently working on a code clean up for this part and will add this to FREVO soon.

    ReplyDelete
  11. Congratulations on your mention in New Scientist (http://www.newscientist.com/article/mg20727723.700-artificial-life-forms-evolve-basic-intelligence.html) and at slashdot.org!

    ReplyDelete
  12. The soccer module for frevotool can be downloaded from http://frevotool.tk -> downloads

    ReplyDelete
  13. Can you make them fake injury? Great article.

    ReplyDelete
  14. Thanks for sharing your info. I really appreciate your efforts and I will be waiting for your further write ups thanks once again.

    ReplyDelete