FetchBench
收藏资源简介:
FetchBench是一个专为评估机器人未知物体抓取能力而设计的数据集,由普林斯顿大学创建。该数据集包含27,500个抓取轨迹,覆盖从日常环境如货架、抽屉中抓取物体的复杂场景。数据集通过程序化场景生成,结合专家抓取演示和地面真实抓取注释,以支持模仿学习方法。FetchBench旨在解决机器人抓取和运动规划在复杂环境中的挑战,提供了一个全面的测试平台,以推动机器人抓取技术的发展。
Created by Princeton University, FetchBench is a dataset tailored specifically for evaluating robotic grasping of unknown objects. It contains 27,500 grasping trajectories, covering complex scenarios of grasping objects in everyday environments such as shelves and drawers. Generated through procedural scene generation, the dataset combines expert grasping demonstrations and ground-truth grasping annotations to support imitation learning methods. FetchBench aims to address the challenges of robotic grasping and motion planning in complex environments, providing a comprehensive testbed to advance the development of robotic grasping technologies.




