Human-Robot Companionships. Mark Neerincx

Transcription

1 Human-Robot Companionships Mark Neerincx

2 TNO and DUT Perceptual and Cognitive Systems Interactive Intelligence International User-Centred Robot R&D Delft Robotics Institute

3 What is a robot? The word robot is derived from the Czech word robota which means serf (labor). In a1920 play of Čapek,, called R.U.R. (Rossum's Universal Robots), it denoted fabricated, artificial human-like beings. A robot is (Russell and Norvig): physical agent manipulates the physical world has sensors has effectors

4 What is a social robot? Merrian-Webster: marked by or passed in pleasant companionship with friends or associates tending to form cooperative and interdependent relationships with others living and breeding in more or less organized communities

5 Misconceptions (should) resemble humans

6 Misconceptions (should) resemble humans can hardly do anything

7 Misconceptions (should) resemble humans can hardly do anything will replace humans

8 Misconceptions (should) resemble humans can hardly do anything will replace humans act (completely) autonomously

9 Evolution of Robots Ishigur, Geminoid HI-1 Androids HRP-4C

10 Evolving Human-Robot Fellowships

11 Human-Robot Collaboration in NIFTi and TRADR Mark Neerincx

12

13 Imperial College London Adaptive cognitive interaction TNO - Delft User modelling University of Hertfordshire Adaptive affective interaction University of Plymouth Long-term memory Aldebaran URBI, Cloud computing and platform Fondazione Centro San Raffaele and Ospidale San Raffaele Evaluation site Vrije Universiteit Brussel A&V perception and production DFKI - Saarbrücken Natural language CNR - Padova Speech recognition and production

14 Introduction Goal: Situated user, context and interaction models, which can evoke both reflective and intentional-strategic robot behaviours empathic, persuasive and engaging robot behaviours for the companion, educator and motivator roles (real-time) dynamic harmonization of these behaviors to the child, the activity and the context, over sustained periods of child-robot interactions Method: Iterative, incremental design and test of the model-based behaviors for the individual user experiences and behaviors addressing the social environment

15 One of the Research Questions How can a personalized adaptive robot help to advance the knowledge, coping behavior and experiences of children with diabetes, by adapting its behavior based on user and context models?

16 Robot as a Quiz Playfellow 22 children played the quiz with another child and with the robot They enjoyed playing with the robot, expressing more emotions Reserved ( introvert ) children were more at ease and confident with the robot Children with age of 7-9 experienced the robot more as a peer than age Girls shared more compliments, whereas boys were more expressive when they were right Recommendations to enrich robot with thinking behaviors, looking up/down, putting hand in front of face, and surprise expressions

17 A Physically Embodied Playfellow Effects of a physical versus virtual robot? Knowledge Attention Trust, Enjoyment, Preferences The physical embodiment of the Nao has a consistent positive effect on attention and is preferred by the children over a virtual embodiment Observations on Collaborative Learning: Children were pride on their performance; when the Nao reflected on their performance and this surpassed Nao s score, they became more competitive and excited

18 The Foundation of Robot Design Knowledge is required Quiz supports learning Zone of Proximal Development Scaffolding Learning Goals Emotion Accommodation Foundation Operational Demands Human Factors Knowledge Technology Innovation

19 Situated Cognitive Engineering Adapt robot state to activity and child state Objective Long-term interaction Mapping of multimodal Nao expressions on affect space Function shapes Interaction Design pattern brings about Effect measures Instrument Emotion contagion Observation

20 Concept map for robot s adaptation

21

22 Concept map for Emotion Expression

23 Emotion Expression

24 Evolutionary development Specification Prototype Simulation Evaluation Foundation Operational Demands Human Factors Knowledge Technology Innovation Test Results

25 Design Rationale Interaction Design Patterns (how) shape Use Case (where, when, who) contextualize Requirements (what) justify Claims (why)

26 Affective Body Postures Scared: yellow, orange eyes. Head, side. Arms, shield eyes. Happy: light eyes. Arms up, straight postures. Angry: red eyes. Head tilted up. Arms in side. Sad: dark blue eyes. Upper body hangs forward. Surprised: light eyes. Upper body moves backwards. Head to side and up. Arms up and sideways.

27 icat vs Nao Emotion recognition Is emotion recognition better when it s expressed by the icat? (facial vs body posture) Does corresponding context increases the recognition rates? (context vs no context) Do recognition rates increase in a second interaction? (1e vs 2nd interaction)

28 Method Fourteen 5th graders, eight and nine year olds. Interacted: twice with Nao (context, no context) twice with icat (context, no context) Second week, same interactions, different context Context: Computer plays a story. Robot plays along and shows emotions No Context: Robot says it will show next emotion Counter-balanced

29 Results Children can recognize both icat and Nao expressions well Overall, no difference in recognition rates, but For Sad, the icat s recognitions were better For Happy, the Nao s recognitions were better Recognition rates higher in Context than in No Context. Second time, higher recognition rates than first time.

30 Design Rationale Interaction Design Patterns (how) shape Use Case (where, when, who) contextualize Requirements (what) justify Claims (why)

31 Method Does the intrinsic motivation of a child to play the Imitation and Quiz (mathematics) game increase when the robot is adaptive on performance? 20 children age 9-10 from school Het Spoor in Zeist, participated in the research, equally divided over 2 groups. Between subject design Measures Intrinsic Motivation Inventory (IMI) Interest/enjoyment and Relatedness Scale 1-7 (helemaal niet mee eens helemaal mee eens) Free-choice period (fcp) Procedure: Three sessions within two weeks Experiment (15 min) fcp (5 min) IMI (15 questions)

32 Results Free-choice period: Adaptive condition significantly higher than nonadaptive condition

33 Results Average level 20 (as was designed) Slow starters played with robot during fcp Average level 24,7 Slow starters played with robot during fcp Children converged to their level over the different sessions

34 Design Rationale Interaction Design Patterns (how) shape Use Case (where, when, who) contextualize Requirements (what) justify Claims (why)

35 Validated Design Rationale Interaction Design Patterns (how) shape Use Case (where, when, who) contextualize Requirements (what) justify Claims (why)

36 Adapt state to activity and child state - within boundaries - Sense child emotion Sense child activity Express emotion Maintain user model Maintain robot model

37 Evaluation Robot with/without adaptive expressiveness Quiz, 18 children (within subjects), age 8-10 Measures: Child s and robot s behaviour Child s perception and preferences Pleasure, acceptance, empathy, trust, emotions and preference Coding scheme for Smiles, laughter, excited bouncing, positive vocalization, frowns, shrugging and sighing, startle, and negative vocalization

38 Results Arousal of robot and child

39 Results Valence of robot and child

40 Results Child s expressiveness per robot adaptive robot nonadaptive robot

41 Results Child s robot preference adaptive robot nonadaptive robot

42 Results Reasons for preferences Nr. Of times given for the non-adaptive robot Understandable Emotion Nr. Of times given for the adaptive robot

43 Design Rationale Interaction Design Patterns (how) shape Use Case (where, when, who) contextualize Requirements (what) justify Claims (why)

44 But, intelligibility should be improved Downside emotion in prosody: worse intelligibility Improve intelligibility measurement and affective prosody

45 Diary Self Management Content and objectives diary scenario based on: Dutch diabetes association Medical experts from hospitals (Rotterdam and Tiel) Focus Group 3 children with diabetes Children do not (like to) use a diary They would like to provide their experiences to a robot

46 Diabetes Diary - Experiment 6 children with diabetes Robot introduction at home with laptop for diary ( and video conference Children had to fill-out the diary daily Every other day they filled out the diary with robot For two weeks

47

48 Example part of diary it was high, so my mood was bad

49 Results

50 Child With robot Without robot Mean Average amount of characters in daily activities logs

51

52 Meander

53 Questions? pitiful when he falls, he may have hurt Charlie has taught me a lot Charlie often says nice things to me Charlie and I are pretty much friends I like him