POMDP Planning for Socially Situated Tasks in HRI

The emerging field of study that is human-robot interaction continues one of the main goals of general robotics research: to create robots that can serve as effective helpmates to humans. Building on the work modeling and acting in physical environments that have led to great successes in navigation and other lower level capabilities, HRI takes on an even more difficult challenge: that of modeling and acting with human participants. As a branch of robotics, a focus on tasks that can be empirically evaluated and approaches that are informed by existing robotics and artificial intelligence will assure that HRI produces robots that are functional as well as compelling. However, there is no reason to expect that the same paradigms, techniques, and approaches that have been so successful in previous robotics applications will also succeed here. Modeling human behavior is a complex and altogether different task, and it is reasonable to expect that new algorithmic approaches will have to be developed to address it. In developing these new approaches, it would behoove us to consider the existing work in related fields such as psychology, even if their theories are not expressed in the form of computational models. A lot of early HRI focused on physical design and social psychology involving people’s perceptions of robots. While those are interesting topics, especially for exploratory research, the overall agenda of HRI research must expand to include building systems that can perform interesting or useful tasks. One good thing about this early focus is that the extreme importance of evaluating systems through their interaction with real humans was immediately established. Simulation makes development easier, but we know that the models of human behavior that we work with are most likely to incomplete for the purposes of rigorous evaluation. People’s subjective impressions of an interaction are important, but the ultimate goal of making robots that are compelling and interpretable to people is so that they may better perform in the roles that they are assigned to. With this in mind, it is important that the tasks we address in our research,whether they are utilitarian or social, have objective measures for success or failure that can be evaluated. Robotics relies on the synthesis of technologies from related fields, and HRI is no different. In order that HRI not be a closed dialog, researchers need to advertise HRI to the greater robotics community as a sub-discipline full of new and interesting application domains. Additionally, it is necessary that we open a dialog between HRI researchers and researchers who develop technologies on which effective