Fast Local Map Construction of Robot Using Semantic Priors

Abstract

When carrying out search or rescue missions, robots often need to explore in an unfamiliar environment. Inspired by the fact that human beings can use their semantic prior knowledge to find objects, robots can also perform search tasks independently according to the semantic relevance of objects in the new environment. In this paper, we propose a scheme that combines SLAM (Simultaneous Localization and Mapping), map construction and object detection to perform the task of target search. By presetting the semantic association rules of common indoor things, the robot can quickly build a local map of the exploration environment on the search path, which is convenient for subsequent use.