Offline Reinforcement Learning: Role of State Aggregation and Trajectory Data