@article{thosar2021rocs,Author = {Thosar, Madhura and Mueller, Christian A. and J├Ąger, Georg and Schleiss, Johannes and Pulugu, Narender and Mallikarjun Chennaboina, Ravi and Rao Jeevangekar, Sai Vivek and Birk, Andreas and Pfingsthorn, Max and Zug, Sebastian},Title = {From Multi-Modal Property Dataset to Robot-Centric Conceptual Knowledge About Household Objects},Journal = {Frontiers in Robotics and AI},Year = {2021},Pages = {87},Month = {04},Doi = {10.3389/frobt.2021.476084},Url = {https://www.frontiersin.org/article/10.3389/frobt.2021.476084},type = {article},Abstract = {Conceptual knowledge about objects is essential for humans, as well as for animals, to interact with their environment. On this basis, the objects can be understood as tools, a selection process can be implemented and their usage can be planned in order to achieve a specific goal. The conceptual knowledge, in this case, is primarily concerned about the physical properties and functional properties observed in the objects. Similarly tool-use applications in robotics require such conceptual knowledge about objects for substitute selection among other purposes. State-of-the-art methods employ a top-down approach where hand-crafted symbolic knowledge, which is defined from a human perspective, is grounded into sensory data afterwards. However, due to different sensing and acting capabilities of robots, a robot's conceptual understanding of objects (e.g., light/heavy) will vary and therefore should be generated from the robot's perspective entirely, which entails robot-centric conceptual knowledge about objects. A similar bottom-up argument has been put forth in cognitive science that humans and animals alike develop conceptual understanding of objects based on their own perceptual experiences with objects. With this goal in mind, we propose an extensible property estimation framework which consists of estimations methods to obtain the quantitative measurements of physical properties (rigidity, weight, etc.) and functional properties (containment, support, etc.) from household objects. This property estimation forms the basis for our second contribution: Generation of robot-centric conceptual knowledge. Our approach employs unsupervised clustering methods to transform numerical property data into symbols, and Bivariate Joint Frequency Distributions and Sample Proportion to generate conceptual knowledge about objects using the robot-centric symbols. A preliminary implementation of the proposed framework is employed to acquire a dataset comprising six physical and four functional properties of 110 household objects. This Robot-Centric dataSet (RoCS) is used to evaluate the framework regarding the property estimation methods and the semantics of the considered properties within the dataset. Furthermore, the dataset includes the derived robot-centric conceptual knowledge using the proposed framework. The application of the conceptual knowledge about objects is then evaluated by examining its usefulness in a tool substitution scenario.}}@COMMENT{Bibtex file generated on }