Experimenting with Robot Emotions: Trials and Tribulations Lilia Moshkina Georgia Institute of Technology Mobile Robot Lab HRI Young Researchers Workshop 2006 Why Robot Emotions? A Wider View: Affect = pertaining to or concerned with emotions A mechanism that allows flexible adaptation to environment and increases survival In Robots Enhances general behavior Transforms into socially interactive companions In People Performs an expressive and communicative function Facilitates understanding and interaction between people and robots TAME framework: Traits, Attitudes, Moods and Emotions Perceptual Input; Current State Behavioral Parameters Motor Vector Behavioral Assemblage TAME Module Active Behaviors Affective State Behavior 1 Behavior … Behavior n Perceptual Module ROBOT Disposition Traits – Big Five •Neuroticism •Extraversion •Openness •Agreeableness •Conscientiousness Affective Attitudes Emotion Component Mood Component Behavior Coordination Environment Exploratory Experimental Study Overall goal: to obtain insights into people’s attitude to affect in robots and to explore ways of assessing affect in robots Study: Robot as a pet and protector scenario Aibo ERS 210A – Sony entertainment robot Longitudinal study – 4 thirty-minute sessions to allow the subjects to bond with the robot A new task for the participant to perform was introduced at each session, and the last session was cumulative. 20 participants – 10 male and 10 female Study Design Independent variable: presence of affect No Affect vs. Affect Conditions Affect condition: personality traits: high on openness, agreeableness, and extraversion emotions: interest, happiness, anger, fear Dependent variables: Pleasantness of interaction Ease of interaction Positive/negative mood Metrics: Introspection (questionnaires) Observation (video analysis) Tasks Follow the ball; kick the ball Participant asks the dog to follow a ball And/or to kick it when close Come to me; Follow me Participant asks the dog to move towards him/her and stop when close or follow “Sick ‘em!” The dog will move to intercept the intruder (Amigobot) Go Play The robot wonders around without requiring user interaction Evaluation Methods Introspection: Negative/positive emotionality (mood) questionnaire Goldberg’s Unipolar Big-Five Markers (personality) Questionnaire Post-study Likert 5-point scale questionnaire: Questions regarding how easy and pleasant it was to interact with the robot; whether or not people could perceive emotions and personality in the robot, and if they could, whether it was more pleasant to interact with a robot possessing emotions and personality. Post-study free-response questionnaire Results Pleasantness of Interaction Overall, participants enjoyed their interaction with the robot (4.27 out of 5) If they thought the robot had emotions and/or personality, they attributed their enjoyment of interaction partly to this feature. They said they enjoyed the interaction better because the robot possessed emotions (4.5) and personality (4.28) 1 2 3 4 5 Questionnaire Response, on scale 1-5 Pleasantness 4.46 4.25 Emotions Personality Results, cont. Ease of Interaction Participants found it fairly easy to interact with the robot (3.98 out of 5, on average) Those in the affective condition found it easier to interact with the robot than those in control condition(p<0.004) May be convoluted due to slight differences in performance Positive/negative Mood Participants experienced less negative mood in the affect condition (p<0.048) Results, cont. 2 There was no significant difference in perception of the robot between the two conditions… however, women were more prone to attribute emotions and personality to the robot regardless of the condition (p<0.043) People were only slightly bored during their interaction, while their enjoyment level was somewhat high. The robot was viewed as more open to experience and more conscientious in the affective condition Results, cont. 3 The majority (15 out of 20) preferred robots with some emotions/personality Reasons for “no” affect: strictly functional applications and ethical issues Would you like to see emotions/personality in robots? 0 1 2 3 4 5 6 7 8 9 10 yes no maybe no answer Number of Participants B A Emotional Non- Emotional Lessons Learned Physical Platform Choice of platform may bias participants’ perception of robot affect Strength of Encoding of Independent Variable Task performance differences Sample size Experiment Design Within-subject vs. between-subject Individual Differences Gender composition Experience with technology Issues to Ponder What Measures can be Used to Evaluate Affective Robotic Behavior? Quantitative vs. Qualitative Performance vs. Perceptions What Metrics can be Applied from Other Disciplines? Questionnaires and standard tests from Psych Observation from behavioral sciences Video Coding Facial expressions Macro vs. micro behaviors Speech utterances New Human-Robot Experiment Robot in a cooperative foraging task Pioneer2-DXE will look for objects in the environment, and humans will place them on top of the robot Independent Variables: Within-subject: presence of affective components: Control (no affect): Emotions only (fear, interest, joy) Emotions, personality and attitudes together Between-subject: Introversion vs. Extraversion (for the multi-affective condition) New Human-Robot Experiment, 2 Dependent Variables and Metrics: Pleasantness of interaction Pre- and post questionnaires Duration and intensity of smiles Positive and negative Mood Implicit Association Test Acceptance Interactivity: verbal interactions and non-verbal behaviors Pre- and post questionnaires Implicit Association Test