ManiSkill

ManiSkill是一种大规模的从演示中学习基准，用于通过视觉输入 (点云和图像) 进行关节式对象操作。ManiSkill通过利用丰富多样的铰接对象集来支持对象级别的变化，并且每个任务都是精心设计的，用于学习对单个对象类别的操作。ManiSkill配备了高质量的演示，以促进从演示中学习的方法并对常见的基线算法进行评估。ManiSkill可以鼓励机器人学习社区探索更多有关学习通用对象操作技能的知识。

ManiSkill is a large-scale learning from demonstration benchmark for articulated object manipulation using visual inputs (point clouds and images). ManiSkill supports object-level variation by leveraging a rich and diverse set of articulated objects, and each task is meticulously designed to learn manipulation skills for a single object category. ManiSkill is equipped with high-quality demonstration datasets to facilitate learning from demonstration methods and enable evaluation of common baseline algorithms. It encourages the robotics learning community to further explore research on learning general-purpose object manipulation skills.

搜集汇总

背景与挑战

背景概述

ManiSkill是一个大规模机器人学习基准数据集，专注于通过视觉输入进行关节式对象操作，支持对象级别变化并提供高质量演示，旨在促进通用对象操作技能的研究。