AnaghaJo/praxis-schema-v0.1

Praxis Schema v0.1 — Human-Hand Towel Pick-and-Place数据集是一个机器人学相关的数据集，包含3个演示片段，展示了以自我为中心的人类操作数据。该数据集的设计目的是为了明确标记失败、恢复和每帧对象状态等可学习信号，而不仅仅是成功的演示。数据集包含丰富的手部语义标签（如左右/主要手、抓取类型、抓取质量等）、全MANO手部状态、对象跟踪信息（如存在性、质心坐标等）、失败注释（如失败模式、失败原因等）以及episode级别的元数据。数据集采用LeRobot-v2布局，包含视频、运动学数据、手部信息、对象跟踪和语义标签等多个关键列。

The Praxis Schema v0.1 — Human-Hand Towel Pick-and-Place dataset is a robotics-related dataset containing 3 demonstration episodes showcasing egocentric human manipulation data. The schema is designed so that failure, recovery, and per-frame object state are explicit, learnable signals — not just successful demonstrations. The dataset includes rich per-hand semantic labels (e.g., left/right/primary, grasp type, grasp quality), full MANO hand state, object tracks (e.g., presence, centroid coordinates), failure annotations (e.g., failure mode, failure reason), and episode-level metadata. The dataset follows the LeRobot-v2 layout, featuring key columns such as video, kinematics, hand information, object tracks, and semantic labels.