MARLeME: A Multi-Agent Reinforcement Learning Model Extraction Library.

Abstract

Multi-Agent Reinforcement Learning (MARL) en-compasses a powerful class of methodologies that have beenapplied in a wide range of fields. An effective way to furtherempower these methodologies is to develop approaches and toolsthat could expand their interpretability and explainability. Inthis work, we introduce MARLeME: a MARL model extractionlibrary, designed to improve explainability of MARL systemsby approximating them with symbolic models. Symbolic modelsoffer a high degree of interpretability, well-defined properties,and verifiable behaviour. Consequently, they can be used toinspect and better understand the underlying MARL systemsand corresponding MARL agents, as well as to replace all/someof the agents that are particularly safety and security critical.In this work, we demonstrate how MARLeME can be appliedto two well-known case studies (Cooperative Navigation andRoboCup Takeaway), using extracted models based on AbstractArgumentation.