HexaJungle: a MARL Simulator to Study the Emergence of Language

Multi-agent reinforcement learning in mixed-motivesettings allows for the study of complex dynamics ofagent interactions. Embodied agents in partially ob-servable environments with the ability to communicatecan share information, agree on strategies, or even lieto each other.In order to study this, we propose a sim-ple environment where we can impose varying levels ofcooperation, communication and competition as pre-requisites to reach an optimal outcome. Welcome tothe jungle.