Abstract

Assume a robot operating in a public space (e.g., a library, a museum) and serving visitors as a companion, a guide or an information stand. To do that, the robot has to interact with humans, which presumes that it actively searches for humans in order to interact with them. This paper addresses the problem how to plan robot's actions in order to maximize the number of such interactions in the case human behavior is not known in advance. We formulate this problem as the exploration/exploitation problem and design several strategies for the robot. The main contribution of the paper than lies in evaluation and comparison of the designed strategies on two datasets. The evaluation shows interesting properties of the strategies, which are discussed.