The continuous development and rise of technology, specifically in the fields of robotics and chatbots, has caught the attention of the Institute für Wirtschaftsinformatik (IWI). With being in possession of the NAO robot, the goal has been set to experiment with their capabilities, in particularly in combination with a chatbot. Therefore, the objective of this Bachelor thesis is to create a prototype, combining the two mentioned components via an intermediate server, developed in Python. Additionally, different chatbots should be evaluated in order to find the one best performing when used within the prototype and thus with the NAO.

Apart from the pre-defined chatbot Dialogflow, a second one had to be selected. Therefore, the two open source chatbot providers Rasa and DeepPavlov were compared in different criteria. Once completed, the intermediate server and various other components of the prototype were developed. To evaluate which chatbot performs best within the protype, several aspects were evaluated such as recognition rates of the chatbots, response times or information extraction. Additional aspects such as the network environment, or a comparison of setup time between the two chatbots was also included in the evaluation phase before making a conclusion.

The output of this Bachelor thesis is a developed prototype consisting of multiple components, to handle a room reservation scenario. These components are an intermediate server developed in Python 3.6, the configuration and integration of the two chatbots Dialogflow and Rasa and the text-to-speech provider Wit.ai, a Choregraphe behaviour to send audio files, and external Python scripts to manage the audio streaming process. The recommendation outcome in terms of chatbots, proposes the usage of the prototype with Dialogflow. This is based on the fact, that Dialogflow triumphs over Rasa in the following three aspects: performance, ease-of-use, and audio file support. It was found that for the Rasa implementation, the speech-to-text provider Wit.ai represents the bottleneck, with requiring long response times and poor input recognition rates. Further findings of this Bachelor thesis, is that the mobile hotspot in which was used to develop the prototype, presents a risk to the performance as the network speed varies heavily.