so101-vla-datasets

SO-101 VLA数据集是一个用于视觉-语言-动作（VLA）学习的机器人操作数据集集合，使用SO-101（SO-ARM101）从动机器人直接收集桌面操作演示，并以LeRobot v3.0格式整理。数据按对象（如立方体、水果）→任务（如拾取放置）→变体（如基础版本、视觉提示覆盖版本）的层次结构组织子文件夹。数据集包含多种模态：两个RGB摄像头的视频、关节状态和动作数据以及自然语言任务指令。它涵盖了多个子集，例如立方体拾取放置、水果拾取放置和基本操作原语（如拾取、推动、放入），并提供了不同变体以支持VLA和ACT等学习任务。数据集规模在10万到100万样本之间，采用Apache-2.0许可证。

The SO-101 VLA dataset is a collection of robot manipulation datasets for vision-language-action (VLA) learning. It is directly collected from the SO-101 (SO-ARM101) mobile robot for desktop operation demonstrations and organized in the LeRobot v3.0 format. The data is structured hierarchically into subfolders by object (e.g., cubes, fruits) → task (e.g., pick and place) → variant (e.g., basic version, visual prompt overlay version). The dataset includes multiple modalities: videos from two RGB cameras, joint state and action data, and natural language task instructions. It covers multiple subsets, such as cube pick and place, fruit pick and place, and basic manipulation primitives (e.g., pick, push, place), with different variants provided to support learning tasks like VLA and ACT. The dataset size ranges from 100,000 to 1,000,000 samples and is licensed under Apache-2.0.