To Kill a Flocking Bird (New Mexico Supercomputing Challenge)

Peter Ahrens and Stephanie Djidjev

Los Alamos High (New Mexico Supercomputing Challenge)

TITLE: To Kill a Flocking Bird

ABSTRACT: This project explores which search techniques work best to optimize the parameters of a flocking model. Flocking is a natural phenomenon of many independent agents (birds) making decisions that lead to the group acting as a whole. The parameters used to control flocking are the angle at which a bird turns to get closer to his neighbors, the angle at which a bird turns to align itself with the rest of the flock and the angle at which a bird turns to get away from his neighbors if he is too close. NetLogo was used to develop an algorithm to judge qualities of a flock, implement the search techniques, run the search techniques and gather the data for comparison. The search techniques used were brute force (a test of all the possible combinations of parameters), genetic algorithms (a random search variant modeling natural selection), bracketing (dividing the search space iteratively), and steepest descent (searching locally and proceeding in the most promising direction to the solution from a random starting point in the search space). To evaluate a flock, a goodness function was created from the following functions: average distance to center, average difference in birds’ distance to center, the average difference in the spacing of each bird to its nearest neighbor, and the average difference the birds’ headings. A visual analysis of the brute force parameter study showed a diagonal gradient through the search space. The other search methods were tested, and compared based on the quality of the flocks produced, the reliability of the search, and the time efficiency. The results showed that the bracketing technique produced the best results.

Back to Swarmfest2010_program