Object-Level Semantic Metric Mapping for Robot Object Search in Home Environment

The environmental map plays a crucial role in facilitating the service robot target object search (TOS). In order to enhance the efficiency of robot TOS and deepen the robot's understanding of the environment, we propose an object-level semantic metric map (OSMM) construction method. First, a multilevel matching approach is proposed by point matching and semantic matching to achieve fusion of camera-based object detection semantic bounding boxes and 2D LiDAR point clouds. Next, for the convenience of map maintenance, a multilevel point cloud clustering method is proposed by using density-based spatial clustering of applications with noise (DBSCAN) and semantic information to remove redundant point clouds. Furthermore, to achieve semantic enhancement, a semantic topological node construction method and an object semantic hierarchical association and reasoning method based on probabilistic model is presented by ontology knowledge. Finally, the comparative experiments and ablation experiments have been implemented in a simulated environment to validate the effectiveness of our mapping method in improving object search efficiency. Additionally, the experiments in real-world scenarios using the TIAGo robot further verify the efficiency of our method.