sEMG Human Dataset
收藏资源简介:
该数据集是由亚马逊FAR实验室、马里兰大学和约翰斯·普金斯大学的研究团队构建的一个多模态力传感数据集,旨在为机器人学习提供富含力信息的示教数据。数据集总时长达10小时,包含同步采集的自中心视频、表面肌电信号、惯性测量单元数据以及指尖力测量值,覆盖了捏取/抓握、拾放、开合和倾倒四种动作类别以及多种手势类型,涉及不同形状、大小和重量的物体。该数据集通过佩戴可穿戴设备在自然交互过程中收集原始信号,并利用薄膜柔性力敏电阻提供精确的指尖力真值,其创建过程融合了硬件设计、数据同步与校准流程。该数据集主要应用于机器人灵巧操作领域,旨在解决从人类示教中学习力敏感操作策略时缺乏接触力监督信息的关键问题,为EMG到力的映射模型训练提供了基础。
This multimodal force-sensing dataset was constructed by research teams from Amazon FAR Lab, University of Maryland, and Johns Hopkins University, aiming to provide force-rich demonstration data for robotic learning. The dataset has a total duration of 10 hours, containing synchronously collected egocentric videos, surface electromyography (sEMG) signals, inertial measurement unit (IMU) data, and fingertip force measurements. It covers four action categories including pinching/grasping, pick-and-place, opening/closing, and pouring, as well as various gesture types, involving objects with different shapes, sizes, and weights. Raw signals were collected during natural human-robot interactions via wearable devices, and precise ground-truth fingertip forces were provided by thin-film flexible force-sensitive resistors. The dataset creation process integrates hardware design, data synchronization, and calibration procedures. This dataset is primarily applied in the field of robotic dexterous manipulation, aiming to address the critical issue of lacking contact force supervision information when learning force-sensitive manipulation strategies from human demonstrations, and provides a foundational resource for training EMG-to-force mapping models.

- 1ForceBand: Learning Forceful Manipulation with sEMG亚马逊·FAR实验室; 马里兰大学; 约翰斯·霍普金斯大学 · 2026年



