RelationalDeepReinforcementLearning--Weintroduceanapproachfordeepreinforcementlearning(RL)thatimprovesupontheefficiency,generalizationcapacity,andinterpretabilityofconventionalapproachesthroughstructuredperceptionandrelationalreasoning.Itusesself-attentiontoiterativelyreasonabouttherelationsbet