Contributors: Marek P. Michalowski, Philipp Michel, Selma Sabanovic, Illah Nourbakhsh
Human social interaction relies on a broad range of abilities: the use of cues such as gaze and gestures, the display of emotions, the adherence to established conventions, the exchange of information through dialogue, etc. We may consider distinguishing between different levels of interaction with the notion that the success of higher-level interaction (in which information or concepts are exchanged) depends on successful interaction at lower levels (i.e. the appropriate use of spatial and temporal conventions). These ``protocols'' are neither necessarily innate nor learned, and they may be unique to human interaction or shared by a larger class of animals.
In order to develop effective human-robot social interaction, whether for the purposes of exploration, tools, or influence, it is necessary to establish basic interactional foundations upon which richer and more sophisticated interactions can be built. We believe that understanding the relationship between appearance, motion, and timing is a fundamental prerequisite for appropriate social interaction. Spatial abilities (the use of bodily motion, attentional cues, and deictic gestures) and temporal properties (speed and rhythmicity) are coupled and must be considered in the same context.
We work with a number of assumptions:
We are designing a robot named Roillo with the long-term goal of examining and testing these ideas. The robot consists of two Reuleaux tetrahedra, two camera eyes, and an ``antenna.'' The body and head can turn and tilt independently, and the antenna can swivel in the range of a half-sphere for deictic gesturing. The robot can hear but not understand human speech, and it can see humans and colorful objects. Our goal is to explore the effective use and combination of spatio-temporally appropriate motion and sound in content-free (i.e. language-free) social interactions.
Generally, we hope to explore our various hypotheses about the basic requirements of form and behavior necessary for peer-to-peer human-robot social interaction. For example, are certain tasks (such as requesting a ball) more easily carried out with separate attentional and deictic gestures than with only one type of gestural cue? Can the natural rhythmicity and melody of vocal interaction be learned to the extent that babbling/content-free conversations can take place?
Another possible use of this robot is as a stationary toy to be situated in playrooms. As an exploratory tool, the robot might be used to observe the differences between how autistic and typically developing children perceive of the robot as a moving object, a living being, and/or a social partner. As a therapeutic device, the robot might be used to encourage and reinforce socially appropriate behaviors such as eye contact and joint attention. We hope to make the robot inexpensive, robust, and deployable in a wide range of situations.