Studying Table-Top Manipulation Tasks: A Robust Framework for Object Tracking in Collaboration.

Table-top object manipulation is a well-established test bed on which to study both basic foundations of general human-robot interaction and more specific collaborative tasks. A prerequisite, both for studies and for actual collaborative or assistive tasks, is the robust perception of any objects involved. This paper presents a real-time capable and ROS-integrated approach, bringing together state-of-the-art detection and tracking algorithms, integrating perceptual cues from multiple cameras and solving detection, sensor fusion and tracking in one framework. The highly scalable framework was tested in a HRI use-case scenario with 25 objects being reliably tracked under significant temporary occlusions. The use-case demonstrates the suitability of the approach when working with multiple objects in small table-top environments and highlights the versatility and range of analysis available with this framework.