Online Object-Oriented Semantic Mapping and Map Updating

Abstract

Creating and maintaining an accurate representation of the environment is an essential capability for every service robot. Especially semantic information is important for household robots acting in indoor environments. In this paper, we present a semantic mapping framework with modular map representations. Our system is capable of online mapping and object updating given object detections from RGB-D data and provides various 2D and 3D representations of the mapped objects. To undo wrong data association, we perform a refinement step when updating object shapes. Furthermore, we maintain a likelihood for each object to deal with false positive and false negative detections and keep the map updated. Our mapping system is highly efficient and achieves a run time of more than 10 Hz. We evaluated our approach in various environments using two different robots, i.e., a HSR by Toyota and a Care-O-Bot-4 by Fraunhofer. As the experimental results demonstrate, our system is able to generate maps that are close to the ground truth and outperforms an existing approach in terms of intersection over union, different distance metrics, and the number of correct object mappings. We plan to publish the code of our system for the final submission.

LTC-Mapping, Enhancing Long-Term Consistency of Object-Oriented Semantic Maps in Robotics

“…In order to evaluate the performance of LTC-Mapping , we have carried out a number of experiments comparing its outcome with the one from a state-of-the-art approach [ 13 ]. The experiments have taken place in different representative environments from the Robot@VirtualHome ecosystem [ 14 ].…”

“…On the other hand, object-oriented methods, which represent objects in the map as individual instances anchored with semantic information, exhibit some advantages that justify their extensive applicability (e.g., a more compact map representation, a more straightforward exploitation, etc.). For example, Dengler et al [ 13 ] present an online semantic mapping method that represents instances of objects as 2D polygons over the geometric map, which are annotated with semantic information (i.e., object class and a likelihood). The available information in the semantic map is updated over time to provide robustness against dynamic changes in the scene and to avoid instance duplication.…”

“…Moreover, to test the adaptability to dynamic environments, we include experiments where objects appear and disappear from those environments. From the obtained results, we demonstrate the suitability of LTC-Mapping , which also exhibits a superior performance than a state-of-the-art alternative [ 13 ], achieving approximately a of object classification accuracy while keeping a low error in modeling their geometry. The method implementation is publicly available at (accessed on 12 July 2022).…”

“…While object-oriented semantic maps bring clear benefits to intelligent robots, two main challenges appear to keep them effective for long-term robot operation: avoiding duplicated object instantiations and coping with dynamic objects [ 11 , 12 , 13 ]. Let us clarify these concepts.…”

3D Semantic Scene Perception using Distributed Smart Edge Sensors

“…Dengler et al [8] proposed an object-centric 2D/3D map representation for real-time service robotics applications, using RGB-D data as input. A geometric segmentation of small objects in the point cloud is obtained via Euclidean clustering.…”