Object Handling of Cognitive Robots Using Deep Leaning Based Object Recognition

In this paper, a methodology of object handling using deep learning based object recognition for cognitive robot is proposed. A cognitive robot using a sentential cognitive system expresses all experienced events as a sentential form and store in a memory to be retrieved for responding to order of human. For the case of an event dealing with objects, a deep learning is used for detecting labels and bounding boxes from the captured data from a RGB-D camera. The segmented 3D information of an object extracted from bounding box and depth data is stored in an object descriptor of the cognitive system for being used for object related conversation. The experimental results show the applicability of the proposed approach to more advanced human robot interaction.