PH2D

该数据集是一个以人为中心的、面向任务的数据集，它与仿人操纵演示相一致，由人类演示收集而来，旨在提高机器人策略学习的效率。该数据集不仅增强了仿人机器人操纵策略的鲁棒性和泛化能力，还与小规模机器人数据共同训练。它包含了多样化的设置，如不同的背景、物体类型和放置位置，其任务专注于为人形机器人开发操纵任务。

This is a human-centric, task-oriented dataset collected from human demonstrations and aligned with human-like manipulation demonstrations, aiming to improve the efficiency of robotic policy learning. This dataset not only enhances the robustness and generalization of humanoid robot manipulation policies, but also is co-trained with small-scale robotic datasets. It features diverse configurations including varied backgrounds, object categories and placement locations, with its tasks dedicated to developing manipulation tasks for humanoid robots.