DROID-3D
收藏资源简介:
DROID-3D是一个大规模的3D机器人操作数据集,包含76K个轨迹(350小时)的高保真交互数据,数据集由天津大学和华为诺亚方舟实验室创建。DROID-3D为3D具身视觉研究提供了宝贵的补充,通过增强原始DROID数据集,加入了高质量的深度图和点云信息。该数据集旨在解决机器人操作任务中训练数据集与实际操作任务之间的领域差距问题,并缺乏能够有效整合3D信息的模型架构问题。
DROID-3D is a large-scale 3D robotic manipulation dataset developed by Tianjin University and Huawei Noah's Ark Lab, which contains 76K trajectories (350 hours) of high-fidelity interactive data. It serves as a valuable supplement to 3D embodied vision research by enhancing the original DROID dataset with high-quality depth maps and point cloud information. This dataset aims to address two core challenges: the domain gap between training datasets and real-world robotic manipulation tasks, and the lack of model architectures that can effectively integrate 3D information.

- 1EmbodiedMAE: A Unified 3D Multi-Modal Representation for Robot Manipulation天津大学, 华为诺亚方舟实验室 · 2025年



