hamza-adnan/visual_distracting_metaworld_with_masks

该数据集包含多个机器人操作任务的模拟环境数据，用于强化学习和计算机视觉研究。每个任务配置（如assembly-v3、basketball-v3、bin-picking-v3等）提供图像观察（包括分心版本）、掩码图像、状态向量、动作向量、奖励信号以及终止和截断标志。数据集还包含预测掩码图像，可能用于分割或注意力建模。数据分为训练集（1,000,000个示例）和测试集（100,000个示例），覆盖了广泛的操作场景，如按钮按压、门开关、抽屉操作、咖啡制作等，适用于多任务学习和迁移学习研究。

This dataset comprises simulation environment data for multiple robotic manipulation tasks, designed for reinforcement learning and computer vision research. Each task configuration (e.g., assembly-v3, basketball-v3, bin-picking-v3) includes image observations (with distracted versions), mask images, state vectors, action vectors, reward signals, termination and truncation flags. The dataset also provides predicted mask images, potentially for segmentation or attention modeling. It is split into training (1,000,000 examples) and test (100,000 examples) sets, covering a wide range of manipulation scenarios such as button pressing, door handling, drawer operations, coffee making, etc., suitable for multi-task and transfer learning studies.