University of Twente Student Theses
Mitigating Unpredictable Robot Actions for Fluent Human-Robot Interaction
Prümm, L.Z.S. (2024) Mitigating Unpredictable Robot Actions for Fluent Human-Robot Interaction.
|
PDF
2MB |
| Abstract: | This master’s thesis explores the mitigation of effects of robot unpredictability on trust from the perspective of ”predictive coding”. The introduction of robots into people’s daily lives holds great potential but also presents challenges, as successful interaction and collaboration require trust in robots. An important aspect for forming trust in a robot is its predictability. However, redictability is not always possible. Sometimes robots will act in ways we do not expect, and this can affect how much we trust them. However, there are ways to lessen these unexpected effects. This report firstly discusses the scientific lense ”predictive coding”, taken in this study. Predictive coding is an approach from neuroscience that describes how people predict the behavior of others. The theory describes the brain as an inference machine. To form expectations, the brain uses a hierarchical forwarding model that compares sensory inputs with what is already known about the situation – internal models. By comparing information from the internal models and the actual sensory input, expectations are formed and rules of behavior are learned. In this report, the theory is for human-robot interactions operationalized as a learning process. In this process, a robot’s predictability becomes more important over time after being consistent in its behavior. As a reason for this we identify the observer’s high confidence in the robot model, which developed through learning the rules and structures of its behavior. Reducing this confidence in the model is identified as our main goal. To achieve that, two time slots are defined: Before the interaction with the robot and before the unpredictable action. Per time slot and based on implications from current related research, a mitigation strategy is developed. This results in two applied strategies: foreshadowing movements before the robot acts unpredictably and informing the observer about changes before interacting with the robot. These strategies are tested in a quantitative study. An experiment is conducted in which participants experience unpredictable robot behavior and evaluate the robot’s predictability and trustworthiness afterwards. The statistical analysis does not provide evidence for the effectiveness of the applied strategies. This does not necessarilymean that the approaches should be completely discarded, but rather that changes to the experimental design should be considered. We eventually discuss what we learned from this study and how we can address the problem better in future research. |
| Item Type: | Essay (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science, 70 social sciences in general, 77 psychology |
| Programme: | Interaction Technology MSc (60030) |
| Link to this item: | https://purl.utwente.nl/essays/99197 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page