Dataset Overview
Definition:AGIBOT WORLD Humanoid Robot Imitation Dataset是面向Humanoid Robots模仿学习的高质量操作TrajectoriesDataset,采集自真实Humanoid Robots在家庭与办公Scenes中的双臂操作过程,同步记录立体广角 RGB-D 视觉、本体感知、腕部六维力与末端执行器Status,可直接用于行为克隆与扩散策略微调。
| Dataset Name | AGIBOT WORLD Humanoid Robot Imitation Dataset |
| Total Data Volume | 10,000+ episodes 完整操作轨迹 |
| Visual Modality | 立体广角 RGB-D,1920×1080 |
| 本体感知 | 关节角/角速度/末端位姿 + 腕部六维力 |
| CoverageScenes | 家庭(收纳、清洁、桌面操作)与办公(整理、搬运、设备操作) |
| 适配格式 | LeRobot 3.0,支持 ACT / Diffusion Policy 微调 |
| 任务标注 | 语言指令 + 任务成功标签 + 轨迹分段 |
| Language | 中文指令为主,可英译 |
Core Data Fields
episode_id— 轨迹唯一标识与场景/任务类型索引rgb_d— 立体广角 RGB 与对齐深度帧序列proprio— 关节角、角速度、末端 6D 位姿本体感知wrench6d— 腕部六维力/力矩高频采样ee_state— 末端执行器开合与夹持状态action_chunk— 动作块序列,可直接用于策略回归
Annotation and Quality Inspection
- 轨迹级人工复核:动作连续性与任务成功标签双审,抽检比例不低于 10%
- 多传感器时间同步校验:视觉/力/本体感知时钟对齐误差可控
- 隐私与合规:采集环境脱敏处理,提供完整授权文件
- 交付前自动化校验:格式完整性、字段缺失率与可读性全量检查
Applicable Scenarios and Delivery
| Applicable Scenarios | Humanoid Robots模仿学习、ACT / Diffusion Policy 策略微调、双臂协同操作研究、力位混合控制探索 |
| Delivery Format | LeRobot 3.0 / HDF5 / JSON 元数据,含标定参数与读取示例 |
| Delivery Cycle | 4-6 Weeks |
| License Type | Commercial Use License + Data De-identification + Complete Authorization Documents |
Datasets in the Same Section
Need Samples or Customization?
The Changsha Langhui Information Technology Co., Ltd. DataAssetsAPI Platform supports flexible customization by scenario, embodiment form and scale。
Inquire NowTactile / Force-Controlled Fine Manipulation DatasetBack to Data Asset CatalogGigaWorld-1 World Model Evaluation Dataset
Need a Customized Embodied AI Data Solution?
Supports flexible combination by scenario, embodiment form, sensor configuration and scale, and provides data sample preview and technical support。