RLA-WM
收藏资源简介:
Maniskill3DWorld 是一个专为3D与多视角研究设计的多模态机器人操作轨迹数据集,源自论文《Learning Visual Feature-Based World Models via Residual Latent Action》的相关工作。数据集的核心内容包括多模态的ManiSkill任务轨迹,具体提供来自7个不同视角的同步数据:RGB图像、深度图和物体掩码。此外,还包含动画化的机器人网格模型和场景的体素点云数据。该数据集适用于机器人学和计算机视觉领域的研究,特别针对视觉特征学习、世界模型构建、3D场景理解和多视角感知等任务。
Maniskill3DWorld is a multimodal robot manipulation trajectory dataset designed for 3D and multi-view research, originating from the work related to the paper Learning Visual Feature-Based World Models via Residual Latent Action. The core content of the dataset consists of multimodal ManiSkill task trajectories, specifically including synchronized data from 7 different perspectives: RGB images, depth maps, and object masks. Additionally, it provides animated robot mesh models and voxel point cloud data of scenes. This dataset is suitable for research in robotics and computer vision, particularly for tasks involving visual feature learning, world model construction, 3D scene understanding, and multi-view perception.




