I. Introduction
For collaborative robots in future household, medical, and education scenarios, learning to manipulate novel articulated objects from human demonstration is a necessary skill. Existing solutions often require multiple human demonstrations [1], additional apparatus [2] [3], or drag-and-teach [4], which could be inefficient or unintuitive. While for OSAL proposed in this research, human only needs to demonstrate by manipulating the object once, which we believe to be the most user-friendly paradigm for general collaborative robots.