Expert-Level Offline Reinforcement Learning Dataset

该数据集包含了100万个用于训练强化学习模型的转换数据，专为模拟环境中的多目标机器人任务（如到达、推动和拾放）而设计。该数据集由完整的剧集组成，没有使用任何不完整的剧集进行训练。它被用于评估决策变压器（Decision Transformer）和结合HER代理的TQC的性能。规模达到了100万个转换。

This dataset contains 1 million transition data for training reinforcement learning models, specifically designed for multi-objective robotic tasks in simulated environments such as reaching, pushing, and pick-and-place. The dataset consists of complete episodes only, with no incomplete episodes used for training. It has been employed to evaluate the performance of Decision Transformer and TQC combined with the HER agent. The total size of the dataset amounts to 1 million transitions.