Learning when to Communicate at Scale in Multiagent Cooperative and Competitive Tasks

Learning when to communicate and doing that effectively is essential in-agent tasks. Recent works show that continuous communication allows training with back-propagation in multi-agent scenarios, but have restricted to fully-cooperative tasks. In this paper, we present Controlled Continuous Communication Model (IC3Net) which has training efficiency than simple continuous communication model, and can applied to semi-cooperative and competitive settings along with the settings. IC3Net controls continuous communication with a gating and uses individualized rewards foreach agent to gain better and scalability while fixing credit assignment issues. Using of tasks including StarCraft BroodWars explore and combat scenarios, we that our network yields improved performance and convergence rates than baselines as the scale increases. Our results convey that IC3Net agents when to communicate based on the scenario and profitability.