TarMAC: Targeted Multi-Agent Communication
Abhishek Das · Theophile Gervet · Joshua Romoff · Dhruv Batra · Devi Parikh · Michael Rabbat · Joelle Pineau

Thu Jun 13th 05:00 -- 05:05 PM @ Hall B

We propose a targeted communication architecture for multi-agent reinforcement learning, where agents learn both \emph{what} messages to send and \emph{whom} to address them to while performing cooperative tasks in partially-observable environments. This targeting behavior is learnt solely from downstream task-specific reward without any communication supervision. We additionally augment this with a multi-round communication approach where agents coordinate via multiple rounds of communication before taking actions in the environment.

We evaluate our approach on a diverse set of cooperative multi-agent tasks, of varying difficulties, with varying number of agents, in a variety of environments ranging from 2D grid layouts of shapes and simulated traffic junctions to 3D indoor environments, and demonstrate the benefits of targeted and multi-round communication. Moreover, we show that the targeted communication strategies learned by agents are interpretable and intuitive.

Finally, we show that our architecture can be easily extended to mixed and competitive environments, leading to improved performance and sample complexity over recent state-of-the-art approaches.

Author Information

Abhishek Das (Georgia Tech)
Theophile Gervet (Carnegie Mellon University)
Joshua Romoff (McGill University)
Dhruv Batra (Georgia Institute of Technology / Facebook AI Research)
Devi Parikh (Georgia Tech & Facebook AI Research)
Mike Rabbat (Facebook)
Joelle Pineau (Facebook)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors