Learning of Intelligent Swarm Behavior

Back to Research
Robert Tjarko Lange

Collective behavior is puzzling and fascinating at the same time. Until now the emergence of such large-scale dynamics has mainly been studied from an evolutionary perspective. In this project, on the other hand, we investigate how normative rewards and top-down learning give rise to intelligent social behavior. More specifically, the goal of the project is to scale Deep Multi-Agent Reinforcement Learning to swarm systems and to provide a set of solutions to the highly non-stationary learning process. By understanding the fundamental mechanisms underlying collective adaptation, we envision the design of artificial shepherds that are able to guide the collective’s inter-temporal decision making process. Thereby, we want to escape sub-optimal equilibria and make the world a better place.