PUB - Publikationen an der Universität Bielefeld

The increasing number of robots in everyday life will lead to more interactions between robots and humans in the future. One interpersonal interaction that takes place in everyday life is the handover of an object. This interaction is a fundamental task for more complex collaborations between two actors. Consequently, in the future, this interaction will also be performed in the context of everyday robot collaborations. To hand over the object, a physical proximity between robot and human is necessary. This proximity can lead to rejection of the interaction by users, as they are unable to perceive the robot or suspect dangers that might be caused by the robot. Therefore, it is necessary to optimize the interaction based on user acceptance and the user’s sense of safety. Overall, three research goals are considered in this thesis. Besides investigating applicationspecific adaptations and extensions for feasible and user-friendly robot-to-human handovers, the focus is also on modeling and implementing anthropomorphic motions for robot-tohuman handovers. The final goal is to evaluate the impact of anthropomorphic motion models on user acceptance and the user’s sense of safety, in particular non-experts. For all three objectives, these will be executed on different types of robots. In order to address this goal, this thesis first presents a robot system that mimics handovers between two humans. For this purpose, results and approaches of related studies are used and adapted for the robotic system. The generation of handover configurations based on specific context parameters is an open question that will be answered in this thesis. To this end, a human-robot study is presented in which test subjects showed robots both appropriate and inappropriate configurations. Based on the statistical results, a model for generating the configurations is proposed. The user’s body size and posture, positioning, and handover object are taken into account. In the development of anthropomorphic motion models for the robot’s arm motion, the focus is on three different conventional motion models. The two basic models are modeled in two different configuration spaces. Thus, the first model is related to the Cartesian trajectory of the end effector and the second model is related to the joint motions of specific joints of the arm. Starting from the models, the recording of the baseline data is a challenge. For this, two studies are shown in which two subjects pass an object to each other. In the process, two different measurement techniques are used to collect the baseline data. Both models show disadvantages in the application based on the given configuration space. As a consequence, a third model is presented that combines the features and supposed advantages of both basic models. During this research objective, different requirements of the models are evaluated subjectively and objectively. The influences of the developed robot system and the modelled movements regarding the feeling of safety and user acceptance are investigated in a human-robot study. The evaluation of the motion models shows that they have no significant influence on the users’ sense of safety. This is applicable to non-experts as well as to experts. Other evaluation criteria, such as intuitiveness, human similarity or user acceptance, are positively influenced by the developed models, as they are mostly performing better compared to reference models. On average, the combined model performed better than the anthropomorphic joint motion model. The anthropomorphic motion model in Cartesian space does not reach an acceptable quality level in the evaluation of human similarity. The evaluation of the other concepts of the robotic system shows different results, where individual hypotheses are rejected against expectations. Further research in the field of robot-to-human handover could deal with the implementation of the interactions in real scenarios. In this context, the kitchen is an exciting field of application, since dangerous objects such as knives and hot utensils are handed over here. These may require extensions to the robot system as well as alternative motion models.

---

Dissertation der Fachhochschule Bielefeld und der Universität Bielefeld