GeorgiaTech/ik-modulation-zarr

该数据集名为IK-Modulation — Segment Zarr，是一个用于人形机器人力条件IK目标调制的分段Zarr数据集，针对Unitree G1人形机器人设计。数据集包含bucket任务（2026-05），其中每个任务将遥操作周期（携带真实IK按压）与人类演示周期配对；通过转移规则将遥操作IK偏移转换为每手XY按压目标，用于训练调制器。数据以Zarr格式存储，包括遥操作和人类演示的最终处理文件，其中遥操作基于IK跟踪误差分段，人类演示基于触觉力（≥1 N，传感器1=左）分段。Zarr布局包含data/state（姿势、触觉、力）、data/action（遥操作IK目标）、data/segment_label（1=左，2=右，3=双手）、data/force_contact和meta/episode_ends。原始数据来源于GeorgiaTech/Lidar-Tactile数据集。该数据集适用于机器人学、触觉感知、人形机器人操作等研究和应用。This dataset, named IK-Modulation — Segment Zarr, is a segmented Zarr dataset for humanoid robot force-conditioned inverse kinematics (IK) target modulation, specifically designed for the Unitree G1 humanoid robot. It includes the bucket task (2026-05), where each task pairs teleoperation cycles (with real IK pressing) with human demonstration cycles; teleoperation IK offsets are converted into per-hand XY pressing targets via transfer rules, which are used to train the modulator. The data is stored in Zarr format, including final processed files for teleoperation and human demonstrations, where teleoperation data is segmented based on IK tracking errors, while human demonstration data is segmented based on tactile forces (≥1 N, Sensor 1 = left hand). The Zarr layout contains data/state (pose, tactile, force), data/action (teleoperation IK targets), data/segment_label (1 = left, 2 = right, 3 = both hands), data/force_contact, and meta/episode_ends. The original data is sourced from the GeorgiaTech/Lidar-Tactile dataset. This dataset is applicable to research and applications in robotics, tactile perception, humanoid robot manipulation and other related fields.