neurips-submission-18302/dmw-submission-18302

该数据集包含多个机器人操作任务的配置，每个配置具有相同的特征结构，包括观察图像、分心观察图像、掩码图像、状态列表、动作列表、奖励值、终止标志、截断标志、预测掩码图像和分心预测掩码图像。数据集分为训练集和测试集，每个配置的训练集有1,000,000个示例，测试集有100,000个示例。数据集用于机器人学习和强化学习任务，但具体任务描述未在README中提供。This dataset includes multiple configurations for robotic manipulation tasks, each with the same feature structure, comprising observation images, distracted observation images, mask images, state lists, action lists, reward values, termination flags, truncation flags, predicted mask images, and distracted predicted mask images. The dataset is split into training and test sets, with each configuration having 1,000,000 examples in the training set and 100,000 examples in the test set. It is intended for robotic learning and reinforcement learning tasks, but specific task descriptions are not provided in the README.