Title: Learning Transferable Cooperative Behavior in Multi-Agent Teams

Abstract: We study the emergence of cooperative behavior and communication protocols in multi-agent teams, for collaboratively accomplishing tasks like coverage control and formation control for swarms. Using graph neural networks to model inter-agent communications, we present state-of-the-art results in a fully decentralized execution framework which assumes agents only have local observability and communication, and show that the learned policies have strong zero-shot generalization to scenarios with additional team members. We also introduce dropout communication, a training methodology which ensures that the learned behavior is robust to communication network glitches and lost packets.
This is an important step towards swarms which can be realistically deployed in the real world without assuming complete prior knowledge or instantaneous, lossless communication at unbounded distances.

Committee:

Katia Sycara (advisor)

Timothy Verstynen

Wenhao Luo