Amazon Robotic Manipulation Benchmark (ARMBench)
收藏资源简介:
ARMBench是由亚马逊机器人创建的大型对象中心数据集,专为仓库环境中的机器人操作设计。该数据集包含超过235,000次拾取和放置活动,涉及超过190,000个独特对象。数据集通过在亚马逊仓库中使用机器人操纵器从内容异质的容器中执行对象分离来收集。ARMBench包含图像、视频和元数据,这些数据捕获了操纵的不同阶段,即拾取前、转移中和放置后。数据集的应用领域包括对象分割、对象识别和缺陷检测,旨在解决仓库自动化中的视觉感知挑战。
ARMBench is a large-scale object-centric dataset developed by Amazon Robotics, tailored specifically for robotic manipulation tasks in warehouse environments. This dataset encompasses over 235,000 pick-and-place operations involving more than 190,000 unique objects. It is collected by deploying robotic manipulators in Amazon warehouses to perform object separation from containers with heterogeneous contents. ARMBench includes images, videos, and metadata that capture distinct stages of robotic manipulation: pre-pick, mid-transfer, and post-placement. Its application scenarios cover object segmentation, object recognition and defect detection, aiming to address visual perception challenges in warehouse automation.




