Body Language for Bots: Adapting a Virtual Robot’s Gestures by Means of an Interactive Evolutionary Algorithm
Robots find application in contexts as diverse as retail, factory work and elderly care. In a social context, it is important that robots can communicate with humans and adapt their behaviour to a changing environment. We raised the question whether we can adapt a robot’s gesture behaviour based on the effectiveness of its communication towards humans. We developed a system that made use of an interactive evolutionary algorithm to develop gestures for a virtual robot. In addition, we investigated if and how interactive evolutionary algorithms can be suitable for the development of gestures for robots. In the experiment we performed (n=16), we found indications that it may well be possible to develop gestures that can be interpreted by other people this way. Additionally, a majority of participants indicated to perceive improvements in the gestures over successive generations of the evolutionary algorithm. However, improvements of the system are necessary, as well as more experiments to confirm the results. To investigate whether gestures that are based on the interaction of multiple people with the system are easier to interpret than gestures developed by an individual, we propose an experiment inwhich gestures are developed by multiple people by means of a transmission chain experiment.
Helena Frijns, "Body Language for Bots: Adapting a Virtual Robot’s Gestures by Means of an Interactive Evolutionary Algorithm", Master's Thesis for the Media Technology programme, Leiden University (The Netherlands), 2017