A Lightweight Semantic Map Construction Method for Robot Target Search

If a robot uses static semantic maps to search targets in dynamic environments, its performance will be unsatisfactory. To solve this problem, a lightweight environment construction method for robot target search tasks is proposed. This novel semantic map can be updated automatically in a dynamic environment, in which unsupervised region division methods are used to establish the attribution of targets to subregions of the environment. Then, a target search strategy is designed to optimize the expected time. Experimental results demonstrated that this semantic map has an excellent real-time and adaptive ability. It does not need a lot of computing resources and storage like 3D point cloud semantic map. It is also convenient for the robot to update automatically online.