Multiagent collaborative task learning through imitation

Abstract

Learning through imitation is a powerful approach for acquiring new behaviors. Imitation-based methods have been successfully applied to a wide range of single agent problems, consistently demonstrating faster learning rates compared to exploration-based approaches such as reinforcement learning. The potential for rapid behavior acquisition from human demonstration makes imitation a promising approach for learning in multiagent systems. In this work, we present results from our single agent demonstration-based learning algorithm, aimed at reducing demonstration demand of a single agent on the teacher over time. We then demonstrate how this approach can be applied to effectively train a complex multiagent task requiring explicit coordination between agents. We believe that this is the first application of demonstration-based learning to simultaneously training distinct policies to multiple agents. We validate our approach with experiments in two complex simulated domains.