LET全尺寸人形机器人真机力觉触觉数据集

本数据集开源首批600余条高质量真机力触觉交互数据，全面覆盖灵巧手在真实操作任务中的多模态感知信息。每条样本同步记录六维力数据（包括抓取、放置等动作中指尖所受的三维力与三维力矩）、高分辨率触觉阵列数据（来自6×12柔性触觉传感器的压力分布矩阵），以及精确对齐的手部关节角度、角速度和末端位姿等运动学信息，实现“视觉-力触-动作”全链路时空同步。数据采集涵盖快递包裹分拣、日化瓶翻面、家庭桌面整理等多种典型操作场景，所有样本均附带详细的语义标注，包括任务类型、物体材质、操作目标及环境上下文，为机器人灵巧操作、触觉感知建模与多模态学习提供坚实的数据基础。

This open-source dataset provides the first batch of over 600 high-quality real-world haptic interaction data samples, comprehensively covering multimodal sensory information of dexterous hands during real-world manipulation tasks. Each sample synchronously records six-dimensional force data (including 3D force and 3D torque exerted on fingertips during actions such as grasping and placing), high-resolution tactile array data (pressure distribution matrix from 6×12 flexible tactile sensors), as well as precisely aligned kinematic information including hand joint angles, angular velocities, and end-effector poses, achieving full spatio-temporal synchronization of the entire "vision-force touch-action" pipeline. The data collection covers various typical manipulation scenarios including express parcel sorting, daily chemical bottle flipping, and home desktop organization. All samples are accompanied by detailed semantic annotations, including task type, object material, operation target, and environmental context, providing a solid data foundation for robotic dexterous manipulation, tactile perception modeling, and multimodal learning.

搜集汇总

背景与挑战

背景概述

该数据集基于全尺寸人形机器人Kuavo 4 Pro采集，提供了涵盖多种真实场景的大规模多任务力觉与触觉数据。它包含指尖触觉阵列、六维力/扭矩等多模态信息，并配有专家标注和完整工具链，旨在支持机器人操作与交互学习。