Flocking is an emergent behavior in which each individual agent follows a simple behavior rule that leads to a group behavior that appears cohesive and coordinated. In our work, we consider how to influence a flock using a set of ad hoc agents. Ad hoc agents are added to the flock and are able to influence the flock to adopt a desired behavior by acting as part of the flock. Specifically, we first examine how the ad hoc agents can behave to quickly orient a flock towards a target heading when given knowledge of, but no direct control over, the behavior of the flock. Then we consider how the ad hoc agents can behave to herd the flock through turns quickly but with minimal agents being separated from the flock as a result of these turns. We introduce an algorithm which the ad hoc agents can use to influence the flock. We also present detailed experimental results for our algorithm, concluding that in this setting, short-term lookahead planning improves significantly upon baseline methods and can be used to herd a flock through turns quickly while maintaining the composition of the flock.

PDF, PS, HTML

In Proceedings of the Ninth International Conference on Swarm Intelligence (ANTS 2014), September 2014.

@inproceedings{ANTS14-katie, title={Influencing a Flock via Ad Hoc Teamwork}, author={Katie Genter and Peter Stone}, booktitle={Proceedings of the Ninth International Conference on Swarm Intelligence (ANTS 2014)}, month={September}, url="http://nn.cs.utexas.edu/?genter:ants14", year={2014} }

Slides (PDF)