Kismet: a robot regulates interaction with its face
At AAAI-98 in Madison (26-30 July 1998) Cynthia Breazeal of the MIT AI Laboratory described the motivational system of the autonomous robot Kismet, which regulates interaction with a person on the model of a caretaker-infant pair: drives, emotions and facial expressions together keep the exchange at an intensity the robot can handle. The robot has active stereo vision with colour cameras in its eyeballs and a microphone on each ear; its moving eyebrows, ears, eyeballs and eyelids show anger, fatigue, fear, disgust, excitement, happiness, interest, sadness and surprise. Early human-robot interaction experiments are reported.
Why it matters
For the first time a robot was made not to perform a task but for a person to interact with face to face, with the person, reading its emotions, tuning the complexity of the interaction to what the robot can learn. From here runs the line of social robots, from AIBO the following year to today’s companions.
The system is built on an agent architecture of transducer processes with activation thresholds: every drive, emotion, behaviour, percept and motor skill is a separate process; the drives maintain a homeostatic regime, the emotions pass activation to the face motor processes and cue the caretaker how to act; the facial features move analogously to human ones and the ears like a dog's. The mouth is absent in the paper, soon to follow. Proceedings of the AAAI Conference on Artificial Intelligence, 15, pages 54-61 by the project's publication list. What the record does not claim: a count of degrees of freedom, which the 1998 paper does not give; the 15 in the face, three for gaze and three for the neck stand on a later MIT project page that describes the robot already with a mouth, four cameras and nine PCs, that is a later configuration; the processing rates and latencies from that page; and the number of participants in the experiments.