Research Archive - Multi Agent RL from Video