Cooperative Object Manipulation Through Multi Agent Offline Reinforcement Learning