etri-robotai-platform-dev/aloha-static-bimanual-real-v1

该数据集是Aloha 14-DoF双手机器人在真实世界中的静态操作基准数据集，包含4种操作任务：电池操作、咖啡操作、杯子打开和魔术贴线程操作。它旨在为VLA（视觉语言动作）、ACT（动作条件变换器）和Diffusion Policy等学习算法提供入门训练数据，并支持通过SimplerEnv ALOHA兼容环境进行评估。数据集排除了ALOHA模拟数据（如插入和transfer_cube_human任务），并从每个来源中精选了前25个episodes，以优化学习效率和循环速度。数据集筛选条件包括任务类别为双手操作、领域为真实世界、机器人系列为aloha，以及episode范围限定为首25个。每个子数据集（battery、coffee、cups_open、thread_velcro）的总episodes数分别为49或50，但仅使用前25个。该数据集由ETRI RobotAI平台自动生成，适用于开发验证，并遵循各来源数据集的原始许可协议。This dataset is a baseline curation for the Aloha 14-DoF bimanual manipulator, focusing on four real-world static manipulation tasks: battery, coffee, cups_open, and thread_velcro. It excludes ALOHA simulation data (such as insertion and transfer_cube_human) and selects the first 25 episodes from each source to prioritize fast learning cycles. Recommended uses include introductory training for VLA (Vision-Language-Action), ACT (Action Chunking Transformer), and Diffusion Policy, with evaluation based on SimplerEnv ALOHA compatibility. Filters specify task categories as bimanual manipulation, domain as real-world, robot family as aloha, and episode range as the first 25. Each sub-dataset has a total of 49 or 50 episodes, with 25 selected for use. The dataset is automatically generated by the ETRI RobotAI platform for development validation and adheres to the original licenses of the source datasets.