Sim-MEES

Sim-MEES是由普渡大学机械工程学院多尺度机器人与自动化实验室创建的大型合成数据集，包含1,550个具有不同难度级别和物理属性的对象，以及1100万个抓取标签，用于移动机械手在杂乱环境中规划抓取。该数据集结合了分析模型和动态模拟整个杂乱环境，以提供准确的抓取标签。数据集内容丰富，包括从ShapeNet、YCB对象、NVIDIA Omniverse资产和DexNet中的对抗性对象中获取的对象。数据集创建过程中，通过结合基于采样的方法、分析模型分析、领域随机化和动态物理模拟，实现了自动化生成准确标记的数据。Sim-MEES旨在为抓取研究提供一个合成基准和参考，适用于不同模态夹具的抓取算法比较和评估，也可用于开发多夹具抓取的仿真到真实数据驱动方法。

Sim-MEES is a large-scale synthetic dataset developed by the Multi-Scale Robotics and Automation Lab, School of Mechanical Engineering, Purdue University. It includes 1,550 objects with varying difficulty levels and physical properties, alongside 11 million grasping labels for grasp planning of mobile manipulators in cluttered environments. This dataset integrates analytical models and dynamic simulations of full cluttered environments to deliver accurate grasping labels. The dataset features a diverse collection of objects sourced from ShapeNet, YCB objects, NVIDIA Omniverse assets, and adversarial objects from DexNet. During its development, automated generation of accurately labeled data was realized by combining sampling-based methods, analytical model analysis, domain randomization, and dynamic physical simulation. Sim-MEES is designed to serve as a synthetic benchmark and reference for grasping research, supporting comparison and evaluation of grasping algorithms across different end-effector modalities, and can also be utilized to develop simulation-to-real data-driven methods for multi-end-effector grasping.