ELIGN: Expectation Alignment as a Multi-Agent Intrinsic Reward