Empirical results from using a comfort level device in human-robot interaction studies