Sociable Machines

1
Sociable Machines

• Cynthia Breazeal
• MIT Media Lab
• Robotic Presence Group

2
Human-Robot Relations
iRobot
Hondas Asimo
Sonys Aibo
Robotic extensions

• Types of relationships
• Face to face with a robot creature
• Embodied, distal interactions through a robot
  avatar
• Augmented physicality through robotic extensions
• Capable machines, untrained users, human
  environment
• Balance human strengths with machine capabilities
• Useful and enjoyable!

3
Robots in your everyday life

• Competence in
• Human engineered environment
• Human social environment
• Untrained users of different
• Age, gender, culture, etc.
• Human centered design
• Appropriate mental model
• Supports what comes naturally
• On the job learning
• Easy to teach
• Long-term relationships
• Acceptance, trust

4
Sociable anthropomorphic robots

• Very complex technology
• Social interface is (ReevesNass)
• Intuitive, natural
• Untrained users
• Humanoid robots are well suited to this
  hypothesis
• Same morphology, sensing
• Share social, communication cues
• HRI meets HCI
• Study how people want to and do interact with
  them.
• Informs design
• Evaluation methods

5
Three research themes
Informed by scientific understanding of
humans And animals
Evaluate robot compatibility with people
HCI
SCIENCE
ROBOTICS AI
Build robots that do real things In the real
world with real people
6
Issues for sociable robots today

• The real-world is
• Complex
• Ever-changing
• Robots have limited abilities compared to people
• Motor skills
• Perceptual abilities
• Mental abilities
• Imbalance in social sophistication
• Yet, social interaction is
• Tightly coupled
• Mutually regulated

7
Early exploration into sociable humanoids

• Set appropriate expectations
• Not human
• Robo-baby
• Use of expressive feedback to regulate
  interaction
• Emotive expressions
• Communicative displays
• Paralinguistic cues
• Use science of natural behavior as a guide
• Start simple and learn, develop

Kismet, MIT AI Lab
8
Socially situated learning A path to more
capable machines?

• Issues for learning systems (robots or otherwise)
• Knowing what matters
• Knowing what action to try
• Evaluating actions
• Correcting errors
• Recognize success
• Structuring learning
• If task is pre-specified, then can do at
  design-time
• If not the case, then what?
• Address issues through structured social
  interactions
• Robots in a benevolent learning environment

9
Learning from the way we teach
10
Social skills that support learning

• Direct visual attention
• Indicates saliency (i.e.what matters)
• Match to human find similar things interesting
• Robot responds to attention directing cues of
  people
• Robot sends feedback to person for focus of
  attention

11
Video of attention system
12
Social skills that support learning

• Recognize communicated reinforcement
• Serves as progress estimator
• Serves as signal for goal attainment
• Robot should recognize affective feedback from
  human
• Robot signal to human that intent was properly
  understood

13
Video of communicated affect
14
Social skills that support learning

• Communicate internal state to human
• Allows human to
• Predict and understand robots behavior
• Tune own behavior to robot
• Improves quality of interaction
• Robot conveys internal state to human in an
  intuitive manner
• Can be used by both to establish better quality
  instruction

15
Communication of internal state
16
Social skills that support learning

• Regulating the interaction
• Provides structure to the interaction
• Interactive games
• Variations on a theme
• Avoid being overwhelmed or under-stimulated
• Turn-taking as cornerstone
• Human interaction
• Human instruction

17
Video of proto-conversations
18
Lessons from Kismet

• Face to face
• In human terms
• Human drive to animate, anthropomorphize
• Importance of gaze
• Social qualities
• Emotive qualities
• Physical interaction
• Being and Feeling in communication
• Expressive feedback is vital
• Entrainment and accommodation
• Mutual regulation
• Being engaged vs. interacting

19
Related, ongoing directions
HRI gaze studies
Smart Puzzle Fruit
HRI DESIGN
SCIENCE
Organic Robots
ENGINEERING
Sensate Silicone Skin
Sociable robots
20
Sociable RobotsStan Winston Studios Media Lab
collaboration

• Next generation sociable robot
• Fully embodied
• Organic look and feel
• Highly expressive
• Socially situated learning

21
Robot Avatars/PerformersStan Winston Studios
Media Lab collaboration

• Symbiotic control
• Puppeteer and single-mind performance
• Human provides content, new interfaces
• Robot local intelligence to perform content
• Physical medium for embodied interactions
• Visual, auditory, tactile
• Mobile
• Shared environment, reference frame
• Physical interactions with world and others

22
Organic RobotsWhat gives a machine a living
presence?

• Organic qualities to make them familiar yet
  distinct
• Intriguing blend between plant and animal
• Silicone skin instead of plastic shells
• Natural and expressive movement, serpentine
• Visual perception of people (faces, movement,
  color)

23
Sensate Synthetic Skin
Perhaps next to the brain, the skin is the most
important of all our organ systems. Ashley
Montagu, Touching The Human Significance of the
Skin, 1986, p.4

• Sensate skin for environmental interactions
• Active perception of material characteristics
  (hard, soft)
• Development of novel conductive silicone sensor
• Neuro-physiological representations

24
Human-Robot Interaction Studies
Controlled studies to better understand the human
side of human-robot interaction

• A series of studies to understand the human
• Focus on the important of gaze in interaction
• Compare physical (robot) verses virtual
  (animation)
• Examine arousal and engagement through autonomic
  responses
• To better understand the advantages and
  limitations of physical vs. animated media